Criteria for inclusion included: (1) Individuals diagnosed with an eating disorder, such as anorexia nervosa, bulimia nervosa, or binge eating disorder according to the DSM-IV or DSM-5 classification systems, or individuals with high-risk factors for the development of eating disorders (e.g., internalization of body ideals, body image concerns, stress, cognitive flexibility deficits and so on) and participating in a written emotional disclosure intervention; (2) Participants of any age, gender or ethnicity; (3) Participants involved in the intervention engaged in expressive writing that focused on emotions and thoughts associated with eating disorder experiences; (4) Studies comparing written emotional disclosure interventions with no intervention/waiting list controls, conventional treatment controls, alternative psychological interventions (e.g., CBT, IPT), placebo, or attentional control interventions; (5) Randomized controlled trials (RCTs) comparing written emotional disclosure intervention to a control group. Clinical trials that are not randomized, including quasi-experimental and controlled trials.

Criteria for exclusion included: (1) Participants with comorbid psychiatric conditions unless data is reported separately for those with eating disorders only; (2) Studies evaluating interventions other than written emotional disclosure (e.g., CBT, IPT, medication); (3) Reviews, case studies, opinion pieces and non-empirical studies. (4) The study population included individuals with unspecified psychiatric or eating disorders, individuals with unspecified risk factors for eating disorders (people with ED-NOS will be excluded), or participants involved in treatment other than written emotional disclosure; (5) Studies that have deviation of intervention mode (e.g., studies that de-emphasized writing about affective experiences of eating disorders, focused on verbal disclosure rather than written disclosure, or did not use writing as a primary intervention component).

A systematic search was conducted across the following databases: PubMed, Medline, Cochrane Library, Web of Science, CINAHL, EBSCO, and Embase. The search included records from the inception of these major databases up to December 31, 2023.

We entered the following search terms in the above database: writing or “expressive writing” or “written disclosure” or “emotional expression” or “emotional disclosure” or “written emotional disclosure” or dairy or self-compassion* and “eating disorder” or “binge eating” or eating disorder* or eating disorders *. Supplementary Table 1 contains specific search formulas for each database.

Titles and abstracts were reviewed independently by two authors (YW and TX). In cases of disagreement, a third author (YT) was consulted. The PRISMA flowchart (Figure 1) summarizes the screening process for identifying, including, and excluding studies at each stage and provides the number of screenings and reasons for exclusion.

Author (YW) extracted the data and the other author (TX) verified the accuracy and completeness of the extracted data. In the face of uncertainty, the two authors discussed with each other and finally reached an agreement. Among the information obtained were: Study characteristics (authors, year of publication, country, and study design). Participant characteristics included sample size, age, gender, eating disorder diagnoses, or risk factors for eating disorders. Intervention details (form of writing intervention, number and duration of writing sessions, control group details, outcome measures and assessment time points). Supplementary Table 2 contains details about the writing contents of the experimental and control groups, time points for assessment, and follow-up rates.

Two authors (YW and XC) independently used the Cochrane Risk of Bias version 2 (RoB 2) tool to evaluate the quality of bias in randomized trials by evaluating the quality of the 13 articles screened. The risk of bias was assessed through five domains: bias in the randomization process, deviations from the intended interventions, missing outcome data, and bias in the measurement of the outcome, with the corresponding entries rated as “Yes, Y,” “Probably Yes, PY,” “Probably No, PN,” “No, N,” or “No Information, NI.” Overall bias would be judged as “Low,” “High” or “Some concerns.”

This paper will provide a narrative summary of the features, interventions, and outcomes of this study. Depending on the heterogeneity of the studies, either fixed or random effects models will be utilized for meta-analysis. Estimate the effect measures of the primary outcome. Studies presenting a high risk of bias will be excluded from the sensitivity analysis. I² will be employed to quantify heterogeneity. In case of significant heterogeneity, subgroup analyses will be conducted to identify potential sources (such as the age of the study population or the follow-up time, etc.). If an adequate amount of research is available, separate meta-analyses will be carried out for subgroups based on the category of eating disorders, the age of the participants, gender, and the form of intervention. Additionally, studies with missing outcome data are evaluated to determine whether appropriate approaches, such as intention-to-treat analyses using inferential methods, are employed to address the missing data. Studies that fail to account for missing data adequately will be regarded as having a greater risk of bias and will be excluded from meta-analysis.

We used the GRADEprofiler 3.6 software (56) to assess the certainty of the body of evidence for outcomes. The tool considers the initial quality of evidence from randomized controlled trials and downgrades or upgrades the quality of evidence based on the rigor of the study design, consistency of the results, precision of the effect sizes, and the potential for publication bias, etc. The quality of evidence output from GRADEprofiler 3.6 is classified into four tiers, namely, high, moderate, low, and very low, and the strength of recommendations is classified into “strong recommendation” and “weak recommendation” to guide clinical decision makers to formulate more precise treatment recommendations based on the understanding of the certainty of the evidence.

The process of study selection involved mainly two authors (YW and TX) independently, and during the process of the evaluation, the two authors discussed and strategized to resolve any uncertainties regarding the eligibility of the studies. Any study on which they disagreed was referred to a third author (YT) to review for inclusion. The first search of the literature base generated 2,847 studies. The remaining 1704 studies were screened for title and abstract after removing 1143 duplicate studies. After title and abstract screening, 1,645 studies were excluded. This meant that 59 studies needed to be assessed by reading the full-text. Following reading the full text, 46 studies were excluded due to non-compliance. 13 studies were ultimately included in the meta-analysis. Figure 1 summarizes our process of study selection.

Among the 13 studies (36–38, 41, 57–65) included in the meta-analysis (Table 1), the majority were from Western European countries (n = 7; 53.8%), with four from the United Kingdom (38, 41, 58, 63), two from Sweden (37, 59), and one from the Netherlands (64); some were from the Americas (n = 4; 30.8%), with three from the United States (36, 57, 62) and one from Canada (61); and two were from Asia (China) (65) and Oceania (New Zealand) (60), respectively. The gender of the majority of the participant population included in the studies was female (n = 1,074; 74.4%), with the majority of the studies targeting women with body image-related concerns (n = 8; 61.5%), and the inclusion criteria for only one study (64) was a confirmed diagnosis of eating disorders. One study (64) did not mention the occupations of the participating population, while the rest of the studies focused on university students, and in the studies that did mention the participating population, the majority of the participants belonged to the white ethnic group and were from the European region.

Two authors (YW and XC) performed a risk assessment of the 13 included studies, the agreement was 80%, and the differences in the assessments were subsequently discussed and revised. The results are shown in the Figures 2, 3. Since few of the included study mentioned information related to allocation concealment, these studies were assessed as “some concerns” when performing the risk of bias assessment of the randomization process. As for the risk of bias, five studies ultimately assessed as “low” (38.5%), five as “some concerns” (38.5%), and three as “high” (23.1%).

Regarding the eating disorder symptoms, eight studies employed diverse eating disorder screening questionnaires (38, 41, 57, 58, 61, 62, 64, 65); Five studies gauged body dissatisfaction by different scales (36, 37, 57, 59, 65); Four studies utilized different versions of the Ideal Body Stereotype Scale to assess the internalization of the ideal body (36, 37, 57, 63). Concerning the measurement of mental health outcomes, two studies utilized the identical anxiety and depression scales (38, 58), and four studies employed distinct negative affect scales (36, 37, 57, 65). Baseline analyses were comparable and data could be merged for meta-analysis.

Publication bias was not assessed because there were fewer than 10 studies in each of the outcome metrics included in this study and the validity of detecting publication bias was limited, which limited our ability to assess the impact of potential publication bias. Therefore, we believe that caution should be maintained in interpreting the results of the study, and future studies need to incorporate a wider range of literature, including unpublished studies, to further enhance the robustness and credibility of the findings.

We evaluated the body of evidence for the outcome indicators involved in the study on the level of evidence, and the results are shown below (Figure 10). This chart shows the credibility of the evidence for eating disorder symptoms, body dissatisfaction, thin ideal internalization, anxiety, depression and negative affect, with eight randomized controlled trials (RCTs) for eating disorder symptoms, four RCTs each for body dissatisfaction and thin ideal internalization, two RCTs each for anxiety and depression, and four RCTs for negative affect. With the exception of negative affect, all of the other studies show a moderate level of evidence credibility.

The results of the meta-analysis showed that WED for people with eating disorders or for those with high risk factors for developing such disorders significantly improved their eating disorder symptoms, which is consistent with the results of the Frayne et al. (40). The potential mechanisms by which written emotional disclosure interventions improve eating disorder symptoms may involve inhibitory and cognitive processes. Individuals’ failure to actively talk about important psychological phenomena can be viewed as a form of active inhibitory behavior acting on the autonomic and central nervous system activity, which is chronic low-level stress (67). Pennebaker et al. (32) showed that WED can help participants to voluntarily disclose emotional or traumatic experiences, reduces inhibitory behaviors, thereby alleviating stress. After the release of stress, individuals experience relief from negative emotions, which enables them to reassess and understand the negative experiences and emotions associated with eating disorders (58). This capacity helps to reduce the excessive focus on and sense of control over food, thereby alleviating the symptoms of eating disorders. Thus, WED interventions can be used as an effective psychotherapeutic modality in the eating disorder community and across a wide range of populations.

Meta-analysis results from this study showed that the WED intervention was able to relieve body dissatisfaction and thin ideal internalization among eating disorder patients and those with high-risk factors developing eating disorders, which is similar to the findings of Stice as well as Rodriguez et al. (68, 69). Melnyk et al. (70) concluded that eating disorders are associated with body image issues. And, improving body image issues can play a role in eating regulation (71). This process may involve emotion and cognition. Research has concluded that WED provides individuals with access to emotional release and helps them deal with negative emotions associated with body image (72). Reassessing and understanding one’s body image by writing about one’s feelings and experiences (73). Self-compassionate expressive writing with a body focus helps people with eating disorders to self-accept and diminish body dissatisfaction by reducing the pursuit of an ideal body image (61).

Meta-analysis showed the high level of heterogeneity, we conducted a heterogeneity analysis. We looked for multiple subgroups, but with unsatisfactory results. After sensitivity analysis, the heterogeneity disappeared after excluding the two studies that had a large impact on the results, and therefore these two studies were considered as possible major sources of heterogeneity. Further analysis revealed that the two excluded studies were of low quality and lacked some specific information (e.g., gender and ethnicity) to be used in the analysis of heterogeneity and could be the source of heterogeneity.

The results of the meta-analysis in this study revealed that the WED intervention had no effect on the levels of anxiety and depression, as well as on the negative affect of the study population. However, research has shown that WED intervention was effective in reducing levels of anxiety and depression in the eating disorder population long after the intervention (38). Another result showed that WED applied to cancer caregivers may improve their adverse conditions of burden, anxiety, and depression (74). It is evident that WED is valid in improving conditions such as anxiety and depression, but the effectiveness may be different due to the different study populations. This may also be attributed to the condition of eating disorder disease, which complicates the negative psychological emotions in this population. Therefore, the role of WED interventions on anxiety, depression, and negative affect in eating-disordered populations cannot be concluded with certainty, more careful consideration is needed.

This meta-analysis has focused on examining the effectiveness of WED interventions for eating disorders (EDs), with results indicating that WED has certain effects in alleviating ED symptoms. However, the realm of emotional expression extends beyond writing, and other forms of emotional expression interventions, such as verbal therapy and art therapy, are also worthy of attention. Verbal disclosure, by providing immediate feedback and interpersonal interaction, may be more conducive for some individuals to deeply understand and process their emotions. Empirical studies have provided evidence to corroborate that verbal disclosure can facilitate emotional expression and cognitive restructuring, which are crucial for the recovery process in eating disorders. Several studies (75, 76) have investigated the role of recovered eating disorder patients as therapists, using verbal self-disclosure in treatment. Results showed that sharing their insights helps patients better understand recovery, reduces shame, and alleviates eating disorder symptoms. This approach effectively encourages patients to self-disclosure, yielding positive therapeutic effects.

Art therapy can enhance an individual’s emotional regulation capabilities and reduce stress and anxiety, both of which are closely related to emotional states associated with ED symptoms. A systematic review conducted by Griffin et al. (77) demonstrated that art therapy enhances self-esteem and alleviates anxiety in individuals with eating disorders by fostering self-expression, increasing self-awareness, and cultivating new perspectives and a sense of pride. Also, the study by Trably et al. (78) showed that art therapy has the potential to enhance specific dimensions of individuals with eating disorders: in particular, music therapy has been shown to significantly alleviate post-meal anxiety, while dance movement therapy can markedly reduce body dissatisfaction.

Nevertheless this meta-analysis did not encompass direct comparisons between these interventions, mainly due to the scarcity of eligible studies and the heterogeneity in methodologies and outcome measurements. Consequently, we advocate for future research to investigate the comparative efficacy of WED against other emotional expression therapies. This exploration is crucial for discerning their respective benefits and suitability in ED treatment. Such analyses would deepen our comprehension of intervention efficacy and uncover individual propensities and reactions to various therapeutic approaches, laying the groundwork for tailored treatment strategies.

The majority of the current research has focused on examining the effects of WED on females and adolescents in Western Europe, so future research could focus more on populations of different ages, genders, geographic areas, and cultural backgrounds. In addition, changes in the environment in which WED is implemented could be considered in the future. Currently, most implementation sites are mainly in laboratories and medical facilities. According to the meta-analysis by Frattaroli, compared to a controlled laboratory setting, WED is more effective at home (79). Also, Gripsrud et al. (80) found in interviews with breast cancer patients that it was easier for them to complete their writing at home.

The frequency, duration, and setting of WED implementation, as well as the instructor who delivered the intervention, were not standardized in the current study. The duration of WED implementation was 15–30 min in most studies, but a few studies fell outside this time range. Most studies chose to conduct WED interventions in the laboratory. The pre-intervention instructors were mostly researchers, with a few adopting peer instruction. Therefore, in the future, consideration needs to be given to the development of a harmonized guideline for clinicians as well as practitioners on the application of WED in the ED population. This guide could focus on describing the frequency and duration of writing, detailing the topics and types of writing, noting the context in which the writing is implemented, considering the interaction of ED population characteristics with writing (e.g., gender, age, geographic region, etc.), and designating the instructor who will implement the intervention.

Several limitations of this meta-analysis need to be mentioned. First, the quality of the included 13 studies was low. Most of the studies only mentioned randomization without elaborating on the specific methods, and very few of them had reference to allocation concealment and specific blinding information. This lack of detail may have some impact on the results of risk of bias assessment. Second, the majority of the study populations were college students and women, the baseline level of eating disorder symptoms in these participants may not be consistent, increasing the risk of selection bias. Third, this study may exist the risk of publication bias. This may be related to the fact that the included studies focused on observing groups of women in Western countries. This limitation is primarily attributed to the existing focus on female patients within the field of eating disorders and the geographical scope of the research. Further attention needs to be paid to individuals from different cultural backgrounds and of different genders in relevant studies in the future, thereby expanding the sample size and validating the consistency of WED’s therapeutic efficacy across various populations. Furthermore, some studies may have obtained more significant results as they used specific scales. Several of the outcome indicators were not scored using a uniform scale, which may have limited our ability to synthesize the results of the studies. And it may be at potential risk of publication bias.

Additionally, some of the outcome indicators were highly heterogeneous and may be associated with sample size, baseline level of the study population, age, gender, cultural background, and duration of follow-up. However, the limited information provided by the included studies resulted in this study not being able to analyze the sources of heterogeneity in depth. In future studies, we will pre-define possible subgroups at the study design stage and include these subgroups (e.g., gender, age, and region) in meta-analyses, even if some information is missing, and try to obtain data from other sources or use alternative indicators. We will also improve the data extraction process, handle missing data more transparently, and select appropriate statistical methods to better understand and account for heterogeneity.

Moreover, the absence of direct comparisons between WED and other emotional expression interventions in our systematic review and meta-analysis. This is primarily due to the diversity of included studies and limitations in data availability. Specifically, the heterogeneity in methodologies, participant characteristics, and outcome measures across different studies complicate direct comparisons. Furthermore, the limited number of studies in the existing literature that directly compare WED with other emotional expression interventions restricts our ability to conduct a comprehensive comparison. Therefore, we recommend that future research should address these limitations to more accurately assess the impact of various intervention measures on eating disorder symptoms.