Prevalence of Eating Disorders in People With Borderline Personality Disorder: A Systematic Review and Meta‐Analysis

ABSTRACT

Objective

The aim of the study was to estimate the prevalence of eating disorders (EDs) across the borderline personality disorder (BPD) population.

Method

After protocol registration (number = CRD42023415721), we systematically searched PubMed/MEDLINE, PsycINFO and Web of Science for articles that allowed the assessment of the proportion of at least one ED in a sample of patients with BPD. Risk of bias was estimated using the Joanna Briggs Institute (JBI) checklist for studies reporting prevalence data, and we used a random‐effects meta‐analytic model. We used the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA 2020) statement.

Results

A total of 34 articles were included, with 20 reports (N = 4107) related to EDs without distinction, and 20 (n = 3901), 20 (n = 4369), seven (n = 766), and six (n = 1773) reports related to anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), and eating disorder not otherwise specified (EDNOS), respectively. The overall frequency of EDs in BPD was estimated to be 29.7% (95% CI = [21.6%, 38.4%]), while the frequencies of AN, BN, BED, and EDNOS were estimated to be 9.98% (95% CI = [5.6%, 15.3%]), 16.3% (95% CI = [12.1%, 21.1%]), 16.3% (95% CI = [6%, 30%]) and 18.8% (95% CI = [10.6%, 28.6%]), respectively. The overall risk of bias was moderate, there was no publication bias, and the certainty levels of our evidence were low.

Discussion

This meta‐analytic review highlights the high frequency of EDs in BPD. It could serve as a basis for exploring the determinants of these comorbidities.

Conclusions

Through this systematic review and meta‐analysis, we found a high prevalence of EDs and their subtypes in patients with BPD.

Summary

• This study highlights the high prevalence of eating disorders in the borderline personality disorder population.

• Bulimia nervosa appears to be more prevalent than the other types of EDs.

• These findings suggest the importance of assessing and monitoring all individuals with a borderline personality disorder for the presence of EDs.

1. Introduction

1.1. Rationale

Borderline personality disorder (BPD) is characterised by a persistent pattern of instability in interpersonal relationships, self‐image and mood, as well as impulsivity, periodic intense anger and feelings of emptiness (Diagnostic and Statistical Manual of Mental Disorders, 2013). Specifically, BPD is associated with disproportionate efforts to avoid abandonment, extreme and oscillating patterns of idealisation and devaluation, identity disturbance, impulsive self‐injurious behaviour, frequent suicidal behaviour, significant mood reactivity, feelings of emptiness, anger control difficulties and possible paranoid ideation or dissociative symptoms. BPD is associated with an increased risk of nonsuicidal self‐injury (NSSI) and suicidal behaviour, with a higher likelihood of premature death by suicide (8%–10% of subjects) compared to the general population (Crocq and Guelfi 2015; Oldham 2006; Paris 2019). The prevalence of BPD is approximately 1.8% in the general population, 6% in primary care settings, 10% among mental health outpatients and 20% in patients admitted to psychiatric wards (Crocq and Guelfi 2015; Gunderson 2008, 2011; Winsper et al. 2020). A variety of comorbid psychiatric disorders are associated with BPD, including substance use disorders (SUDs), mood disorders, anxiety disorders and eating disorders (EDs) (Crocq and Guelfi 2015; Leichsenring et al. 2024).

With a lifetime prevalence of approximately 8.4% for females and 2.2% for males (Galmiche et al. 2019), EDs can be defined as severe and persistent disturbances in eating behaviour that result in pathological eating patterns or food intake with detrimental effects on psychological wellbeing, physical health and social functioning (Diagnostic and statistical manual of mental disorders, 2013). Types of EDs include anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), avoidant restrictive food intake disorder (ARFID), other specified feeding and eating disorder (OSFED), pica, rumination disorder, eating disorder not otherwise specified (EDNOS) and unspecified feeding or eating disorder (UFED) (Diagnostic and Statistical Manual of Mental Disorders, 2013). For example, AN involves restricting energy intake to meet needs, with subsequent low body weight, disproportionate fear of weight gain, self‐image dysfunction and persistent lack of importance of low body weight. AN can be of restricting or binge‐eating/purging type, depending on the engagement in bingeing or purging. On the other hand, BN is characterised by recurrent episodes (at least once a week for 3 months) of binge eating, frequent inappropriate compensatory behaviours such as self‐induced vomiting. The frequency of AN in women and men is nearly 1.4% and 0.2%, respectively, while the prevalence of BN is estimated to be 1.9% in women and 0.6% in the male population (Galmiche et al. 2019). EDs are often associated with psychiatric and somatic comorbidities (Udo and Grilo 2019), which contribute to impaired quality of life and an increased mortality (Pohjolainen et al. 2016; Smink et al. 2012). Mental health comorbidities include personality disorders, especially BPD (Bruce and Steiger 2005; Campbell 2009; Chen et al. 2011; Gazzillo et al. 2013; Helverskov et al. 2010; Kendall and Clarkin 1992; Martinussen et al. 2017).

According to previous meta‐analytic reviews, the prevalence of BPD in patients with EDs, AN, BN, EDNOS and BED is estimated to be 22%, 19%, 25%, 12% and 10%, respectively (Friborg et al. 2014; Martinussen et al. 2017). The co‐occurrence of personality disorders and EDs negatively affects the long‐term development of EDs and reduces the benefits of ED treatment (Bruce and Steiger 2005; Campbell 2009; Chen et al. 2011; Gazzillo et al. 2013; Helverskov et al. 2010; Kendall and Clarkin 1992; Martinussen et al. 2017). The presence of one or more personality disorders in individuals with an ED, particularly those with BPD and BN, may weaken adherence to treatment and prognosis (D. B. Herzog and Sacks 1993; T. Herzog et al. 1991; Johnson et al. 1990; Newton 2019; Rorty and Yager 1996; Skodol et al. 1993; Steiger et al. 1994). A previously published literature review on BPD comorbidities reported that the frequencies of AN, BN and EDNOS ranged from 0% to 21% (median = 6%), 3% to 26% (median = 10%) and 14% to 26% (median = 22%), respectively (Shah and Zanarini 2018). EDs negatively influence the development of BPD. Compared to the general population, an increased number of suicide attempts in BPD patients with BN and a higher risk of self‐harm behaviour among BPD patients with AN have been reported (Chen et al. 2009; Danon et al. 2022; Skodol et al. 1993).

In summary, the significant mortality and morbidity associated with BPD, together with the co‐occurrence of EDs and BPD, suggest the importance of quantifying the prevalence of EDs in populations of patients meeting criteria for BPD. Considering this fact, we planned a synthesis of the current literature. Such a project could strengthen the current knowledge on the interaction between BPD and EDs, but also serve as a rationale to explore the determining factors of this clinically relevant comorbidity.

1.2. Objective

The aim of this study was to conduct a meta‐analysis to assess the prevalence of EDs in patients meeting criteria for BPD.

2. Methods

2.1. Protocol and Registration

Before conducting this meta‐analytic review, we searched existing reports to identify any recent review pertaining to our topic. After this exploration, we designed a review protocol, which we registered to the International Prospective Register of Systematic Reviews (PROSPERO) under the following number: CRD42023415721. We based our synthesis on the Preferred Reporting Items for Systematic Reviews and Meta‐analyses (Page et al. 2021).

2.2. Eligibility Criteria

Our review focused on articles published in French or English, without temporal or sociodemographic limitations, that included observational studies (cross‐sectional, case–control or cohort studies) reporting the prevalence of at least one ED (the studied condition), in a sample of patients meeting criteria for BPD (identified by validated assessment measures such as reference‐based scale, score or questionnaire). We also included studies with enough information to assess the prevalence of EDs in the sample and targeted all types of EDs (AN, BN, BED, ARFID, OSFED, pica, rumination disorder, EDNOS and UFED). EDNOS is the category in DSM‐IV dedicated to EDs of clinical severity that do not meet diagnostic criteria for either one of the two EDs recognised in DSM‐IV, AN and BN (Fairburn and Bohn 2005). We excluded case reports, case studies, observational studies with fewer than 30 participants, letters to the editor, commentaries, review articles, duplicates, the least recent studies if many studies were based on the same population and articles for which full text was not available.

2.3. Information Sources and Search Strategies

We searched PubMed/MEDLINE, PsycINFO and Web of Science for relevant articles. Our search strategies were designed by associating groups of words/expressions related to BPD and EDs. We first applied these strategies on 15 March 2023 and updated searches on 28 June 2025. The search strategies applied to each database are shown in Table 1. After applying the search strategies in selected databases, the respective resulting files were merged as a common project in a platform dedicated to support knowledge synthesis projects, called Rayyan.

TABLE 1.

Search strategies for each database.

Database	Search strategy
PubMed/MEDLINE	(‘borderline personality disorder’[Title/Abstract]) AND (‘eating disorder*’[Title/Abstract] OR ‘anorexi*’[Title/Abstract] OR ‘bulimi*’[Title/Abstract] OR ‘binge‐eating’[Title/Abstract] OR ‘avoidant restrictive food intake disorder’ [Title/Abstract] OR ‘other specified feeding and eating disorder’[Title/Abstract] OR ‘pica’[Title/Abstract] OR ‘rumination disorder’[Title/Abstract] OR ‘eating disorder not otherwise specified’[Title/Abstract] OR ‘unspecified feeding or eating disorder’[Title/Abstract])
Web of Science	((‘Borderline Personality Disorder’) AND (‘Eating Disorder*’ OR ‘Anorexi*’ OR ‘Bulimi*’ OR ‘Binge‐Eating’ OR ‘Avoidant Restrictive Food Intake Disorder’ OR ‘Other Specified Feeding and Eating Disorder’ OR ‘pica’ OR ‘Rumination Disorder’ OR ‘Eating Disorder Not Otherwise Specified’ OR ‘Unspecified Feeding or Eating Disorder’))
PsycINFO	(‘borderline personality disorder’ and (‘eating disorder*’ or ‘anorexi*’ or ‘bulimi*’ or ‘binge‐eating’ or ‘avoidant restrictive food intake disorder’ or ‘other specified feeding and eating disorder’ or ‘pica’ or ‘rumination disorder’ or ‘eating disorder not otherwise specified’ or ‘unspecified feeding or eating disorder’)).ab.

2.4. Selection Process and Data Collection

The study selection process was carried out by two independent assessors (TP and FTE) in two steps, taking into account eligibility criteria. The first step was based on titles and abstracts, and for the second one, we retrieved and analysed full texts of the articles that remained after the initial selection. We resolved discrepancies by consensual decisions or through the intervention of a third assessor (FN). At the end of the selection process, we extracted data using a predefined form that included the following elements:

1. Study characteristics including first author name, publication year, study design, study location (country and region) and data collection period;

2. Participants characteristics, including the percentage of females, whether or not adolescents were included, mean age, sample size and types of EDs studied;

3. Information on the instrument/tool used to identify (diagnose) BPD and EDs among people with BPD;

4. Prevalence data or data that allowed the prevalence of EDs to be calculated in people with BPD.

In addition, we sought information that would allow us to assess the methodological quality of included studies.

2.5. Study Risk of Bias Assessment

The risk of bias of included studies was assessed independently by two investigators (TP and FTE) using the Joanna Briggs Institute (JBI) critical appraisal checklist for studies reporting prevalence data (Moola et al. 2015). This checklist is designed to assess the following nine aspects: (a) sample frame adequacy, (b) recruitment method, (c) sample size, (d) participants and the setting, (e) coverage, (f) diagnostic methods, (g) the reliability and standardisation of measurements, (h) statistical analysis and (i) the response rate. Specific questions correspond to each aspect/item, with ‘yes,’ ‘no,’ or ‘unclear’ as response options. The total amount of ‘yes’ allow to be classified as low risk of bias (≥ 70%), moderate risk of bias (≥50% and <70%) and high risk of bias (<50%).

2.6. Effect Measures

We analysed collected data using the R statistical software (Version 4.0.3, the R Foundation for Statistical Computing, Vienna, Austria). The prevalence of each ED was estimated from each study using the ratio between the total number of the concerned ED to the total number of individuals with BPD, and we expressed this prevalence as proportions. To pool data of interest (for EDs in general and for each subtype), we performed a random‐effects meta‐analysis with a 95% confidence interval (CI). An arcsine square root transformation (Freeman–Tukey) helped us lower the potential effect of studies with significantly high or low prevalence data (Borges Migliavaca et al. 2020; Freeman and Tukey 1950). To illustrate our findings we used forest plots, we assessed heterogeneity using the Cochrane's Q statistic (Higgins and Thompson 2002) and measured by I ² values (and CIs) with the following thresholds: i ² < 0.25 = low heterogeneity, I ² between 0.25 and 0.5 = moderate heterogeneity, I ² > 0.5 = high heterogeneity. For our main results, as an additional measure of heterogeneity, we used H ² (H‐squared) values. Note that a H ² = 1 implies no heterogeneity, whereas values > 2 are interpreted as significant heterogeneity.

In addition to the Freeman‐Tukey transformation, we estimated the proportions of the target conditions (unspecified EDs and ED subtypes) in people with BPD using generalised linear mixed models (GLMMs). Specifically, we calculated each estimated proportion (with 95% CIs) using the fixed effect estimates, Z‐values and standard errors.

We performed subgroup analyses considering regional locations of each included study, genders, population classes according to age, ED assessment tools and methods used to explore the BPD. We also performed a meta‐regression with mean age, male/female ratio and sample size as potential determinants. We concluded significant results for a p value < 0.05.

2.7. Reporting Bias Assessment

The publication bias was appraised through a visual inspection of funnel plots, reinforced by the Egger's test for analyses including at least 10 studies (Egger et al. 1997). We concluded significant results for a p value < 0.1.

2.8. Certainty of Evidence

The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach allowed us to assess the certainty of evidence (Balshem et al. 2011; GRADE handbook, s. d.). This approach is based on five axes including the risk of bias (of selected studies), the heterogeneity, the indirectness (generalizability), the imprecision and the publication bias. After appraisal of these axes, the certainty of the evidence certainty is characterised as high, moderate, low or very low (Balshem et al. 2011; GRADE handbook, s. d.).

2.9. Synthesis Methods

Our results were summarised using text, tables and graphs. We addressed the following points: a) synthesis of the article selection process (text and graphical presentation), b) synthesis of the main characteristics of the studies (text and tables), c) synthesis of the risk of bias assessment (text and tables) and d) synthesis of the pooled prevalence estimates including subgroup analyses, publication bias assessment and evidence certainty assessment (text, tables, forest plots and funnel plots).

3. Results

3.1. Study Selection

After applying search strategies to predefined databases, we obtained 1321 articles from which 636 duplicate files were deleted. From the remaining articles (685), based on the evaluation of the titles/abstracts, we retained 73 papers and scrutinised their full texts considering our eligibility criteria. In these remaining 73 articles, we excluded three articles because of the language (two in German and one in Spanish), seven articles due to the lack of prevalence data and six articles due to the wrong publication type. In addition, three studies were not retained because they used self‐diagnosis (without the use of a scale, a score or a questionnaire). For the other excluded papers, one study was an extension of another study already included, two studies had a sample size of less than 30, and 20 studies were outside the topic of interest. Thirty‐one studies were initially selected, and after refining the search strategy, 34 studies were ultimately included in our systematic review and meta‐analysis, as displayed in the PRISMA‐based flowchart of study selection (see Figure 1).

FIGURE 1.

Flow chart of study selection.

Of the 34 included articles, 20 reports were related to EDs without distinction (4107 participants), while AN, BN, BED and EDNOS concerned 20 (n = 3901), 20 (n = 4369), seven (n = 766) and six (n = 1773) reports, respectively. There were no papers reporting relevant data on ARFID, OSFED, pica, rumination disorder, EDNOS and UFED.

3.2. Characteristics of the Included Studies

The 34 papers we examined for our meta‐analytic review were published between 1993 and 2025, with study periods ranging from 1986 to 2020. Of the 20 studies that assessed the prevalence of EDs across BPD, 18 had a cross‐sectional design, two included adolescents (with or without adults), and three included only women. Studies addressing the general prevalence of EDs were conducted in the United States of America or USA (n = 8), in Spain (n = 4), in Germany (n = 1), in China (n = 1), in Iran (n = 1), in Japan (n = 1), in Norway (n = 1), in the Netherlands (n = 1), in Poland (n = 1) and in Taiwan (n = 1). The identification of EDs was based on the DSM‐IV in nine studies and the DSM‐III‐R in four studies. The MINI was used in three studies. Regarding BPD, the DSM‐IV criteria were used in 11 studies, whereas the DSM‐III‐R was used in three studies.

All the studies that assessed AN were cross‐sectional, four implied adolescents (with or without adults), and four targeted exclusively women. As studies' locations, we had the USA (n = 8), Germany (n = 4), Norway (n = 2), Japan (n = 1), Italy (n = 1), Iran (n = 1), Spain (n = 1) and China (n = 1). One study was multisite. AN was mainly assessed with the DSM‐IV (n = 7), the DSM‐III‐R (n = 6) and the MINI (n = 4). BPD was predominantly appraised using the DSM‐IV (n = 13) and the DSM‐III‐R (n = 5).

Regarding BN, all studies were cross‐sectional, three included adolescents, and five exclusively focused on women. The studies' countries were the USA (n = 8), Germany (n = 3), Norway (n = 3), Japan (n = 1), Italy (n = 1), Iran (n = 1), Spain (n = 1), China (n = 1) and one study concerned more than one site. The identification of BN was mostly performed using the DSM‐IV (n = 6), the DSM‐III‐R (n = 6) and the MINI (n = 6), while BPD was mainly evaluated through the DSM‐IV (n = 13) and the DSM‐III‐R (n = 5).

All studies on BED and EDNOS were cross‐sectional and exclusively concerned adults except for one research work on EDNOS. Studies on BED were located in Germany (n = 2), the USA (n = 1), Iran (n = 1), India (n = 1) and Egypt (n = 1), while studies related to EDNOS were conducted in the USA (n = 4), Norway (n = 1) and China (n = 1). The most used tool for EDNOS assessment was the DSM‐III‐R, while the most utilised one for BED was the DSM‐IV. Tables 2 and 3 show the main characteristics of the studies and participants included in our review.

TABLE 2.

Characteristics of the included studies and participants.

Article	Study design	Study period	Country	Study setting	Population	Gender	Percentage of females
Banzhaf et al. (2012)	Cross‐sectional	2004–2009	Germany	Inpatient treatment program	Adults	Males and females	66%
Barral et al. (2016)	Cross‐sectional	Missing data	Spain	Outpatient from drug clinic centre	Adults	Males and females	44%
Black et al. (2007)	Cross‐sectional	Missing data	USA	Subjects (prisoners) selected from the list of incoming offenders newly committed	Adults	Males and females	18.5%
Casellas‐Pujol et al. (2024)	Cohort Prospective	2001–2021	Spain	Population from outpatient unit	Adults	Males and females	88.1%
Chen et al. (2009)	Cross‐sectional	Missing data	USA	Population from randomized controlled trials selecting for suicidal BPD or substance‐dependent BPD	Adults	Females	100%
Cortés‐García et al. (2021)	Cross‐sectional	2008–2016	Norway	Adolescent inpatients recruited as part of a larger study from a private psychiatric hospital	Adolescents +/− adults	Females	100%
Del Pozo et al. (2018)	Cross‐sectional	2013	Germany	Adults who consecutively underwent outpatient or inpatient psychotherapeutic treatment from clinics for psychosomatic illnesses and at a university psychotherapy outpatient department	Adults	Females	80%
Dulit et al. (1994)	Cross‐sectional	1990–1992	USA	Inpatients consecutively admitted to a clinic	Adults	Males and females	79%
Farrés et al. (2018)	Cross‐sectional	Missing data	Spain	Outpatients who were attended in the borderline personality unit of the hospital	Adults	Males and females	87.5%
Frías et al. (2017)	Cross‐sectional	2015–2017	Spain	Adult outpatient of mental health center	Adults	Males and females	91%
Johnson et al. (2003)	Cross‐sectional	Missing data	USA	Participants is part of a larger study of the longitudinal course of PDs, the collaborative longitudinal personality disorders study (CLPS).	Adults	Males and females	73%
Kahl et al. (2012)	Cross‐sectional	Missing data	Germany	Adults' psychiatric inpatients	Adults	Males and females	89.6%
Khosravi et al. (2020)	Cross‐sectional	2018–2019	Iran	Participants selected by systematic random sampling among the persons who referred to Baharan Psychiatric Hospital	Adults	Males and females	BPD patients without ED: 55% BPD patients with ED: 64%
Kockler et al. (2017)	Case–control	Missing data	Germany	All‐female sample out/inpatients occurred at the Central Institute of Mental Health Mannheim and at the psychosomatic clinic	Adults	Females	100%
Kot et al. (2023)	Cross‐sectional	Missing data	Poland	Participants hospitalized at a specialized inpatient therapeutic ward	Adults	Females	100%
López‐Villatoro et al. (2025)	Cross‐sectional	Missing data	Spain	Patients were recruited from a personality disorders day hospital	Adults	Males and Females	86.8%
Marino et al. (2000)	Cross‐sectional	1992–1995	USA	Inpatients	Adults	Females	100%
Miller et al. (2019)	Cross‐sectional	2008–2010	USA	Adolescent patients from a psychiatric inpatient or partial hospitalisation unit	Adolescents +/− adults	Males and females	Missing data
Moriya et al. (1993)	Cross‐sectional	1986–1989	Japan	Outpatients at a university hospital	Adults	Males and females	100%
Perugi et al. (2011)	Cross‐sectional	Missing data	Italy	Participants came from a variety of sources, about equally divided between self‐referrals, referrals from general practitioners and various medical specialists and psychiatrist	Adolescents +/− adults	Males and females	87%
Reas et al. (2013) (Reas _ 1)	Cross‐sectional	1993–2009	Norway	Participants are from consecutive and first admissions to 16 different treatment units in the Norwegian Network of Personality‐Focused Treatment Program	Adults	Males and females	Missing data
Reas et al. (2016) (Reas _ 2)	Cross‐sectional	1996–2009	Norway	Participants are from consecutive and first admissions to 16 different treatment units in the Norwegian Network of Personality‐Focused Treatment Program	Adults	Females	100%
Reas et al. (2015) (Reas _ 3)	Cross‐sectional	1996–2009	Norway	Participants are from consecutive and first admissions to 16 different treatment units in the Norwegian Network of Personality‐Focused Treatment Program	Adults	Females	100%
Shaker et al. (2022)	Cross‐sectional	2019–2020	Egypt	Outpatient clinic	Adults	Males and females	68%
Shen et al. (2018)	Case–control	2003–2006	Taiwan	Data extracted from the LHID 2005 (total general population), participants included were newly diagnosed with BPD by a psychiatrist	Adolescents +/− adults	Males and females	65%
Shenoy et al. (2019)	Cross‐sectional	2017	India	Students	Adults	Males and females	52%
Silberschmidt et al. (2015)	Cross‐sectional	Missing data	Multisite	Population recruited into the study from 107 sites in 17 countries (Argentina, Belgium, Chile, France, Germany, Italy, Norway, Peru, Poland, Portugal, Romania, Spain, Sweden, Turkey, UK, US and Venezuela). Sites included academic centres, hospitals, freestanding clinics and clinical trial research facilities. All subjects were outpatient	Adults	Males and females	72%
Slotema et al. (2018)	Cross‐sectional	2012–2015	Netherlands	Outpatient from a clinic specialised in the treatment of personality disorders	Adults	Females	100%
Tadić et al. (2009)	Cross‐sectional	1999–2005	Germany	Inpatients	Adults	Males and females	69%
Wong et al. (2010)	Cohort Prospective	2007–2008	China	Inpatients Chinese ethnicity (accident and emergency department) within 48 h of admission	Adults	Males and females	83%
Zanarini et al. (2004) (Zanarini _ 1)	Cohort Prospective	1992–1995	USA	Inpatients	Adults	Males and females	77%
Zanarini et al. (2021) (Zanarini _ 2)	Cross‐sectional	2007–2012	USA	Inpatients	Adolescents +/− adults	Males and females	Adolescents: 94% Adults: 80%
Zanarini et al. (1998) (Zanarini _ 3)	Cross‐sectional	1991–1995	USA	Inpatients	Adults	Males and females	78%
Zlotnick et al. (2002)	Cross‐sectional	Missing data	USA	Outpatient private practice that is part of a general hospital	Adults	Males and females	70%

TABLE 3.

Other characteristics of the participants.

Article	Tool or manual used for EDs identification	Tool or manual for BPD identification	Type of EDs assessed	Sample size	Mean age (SD) in years	Age range in years
Banzhaf et al. (2012)	MINI	DSM‐IV SCID II	AN, BN, BED	170	Missing data	Missing data
Barral et al. (2016)	DSM‐IV SCID‐I	DSM‐IV SCID‐II	EDs without specification	128	34 (7.75)	Missing data
Black et al. (2007)	MINI	DSM‐IV	EDs without specification, AN, BN	65	29.5 (7.3)	Missing data
Casellas‐Pujol et al. (2024)	Clinical interview	DSM‐IV SCID II DIB‐R	EDs without specification	620	29.9 (7.9)	Missing data
Chen et al. (2009)	DSM‐IV SCID‐I	DSM‐IV IPDE SCID‐II	EDs without specification, AN, BN, BED	135	30 (7.6)	18–45
Cortés‐García et al. (2021)	NIMH DISC‐IV	DSM‐IV CI‐BPD	AN	149	15.3 (1.17)	12–17
Del Pozo et al. (2018)	DSM‐IV SCID‐I	DSM‐IV SCID‐II	AN	32	Women and men: 30.3 (9.02) Women: 29 (7.9)	Missing data
Dulit et al. (1994)	DSM‐III‐R SCID‐I	DSM‐III‐R SCID‐II	AN, BN	124	Missing data	18–60
Farrés et al. (2018)	DSM‐IV	DSM‐IV MSI‐BPD SCID‐II DIB‐R	EDs without specification	118	29.8 (7.76)	18–50
Frías et al. (2017)	DSM‐IV SCID‐I/ P	DSM‐IV SCID‐II	EDs without specification	102	35.99 (11.09)	18–65
Johnson et al. (2003)	DSM‐IV	DSM‐IV DIPD‐IV	EDs without specification	240	Women: 31.55 (8.14) Men: 32.78 (7.57)	18–45
Kahl et al. (2012)	DSM‐IV SCID‐I	DSM‐IV SCID‐II	AN, BN, BED	135	31.9 (9.8)	18–56
Khosravi et al. (2020)	EAT – 26	DSM‐5 SCID5‐PD BPI	EDs without specification, AN, BN, BED	110	Missing data	18–35
Kockler et al. (2017)	DSM‐IV SCID‐I	DSM‐IV SCID‐II IPDE	BN	43	26.72 (7.07)	18–48
Kot et al. (2023)	Missing data	DSM‐V SCID‐II PD	EDs without specification	100	24.19 (5.41)	18–45
López‐Villatoro et al. (2025)	MINI	DSM‐V SCID‐II	AN, BN	141	29.31 (9.26)	Missing data
Marino et al. (2000)	DSM‐III‐R SCID‐I	DSM‐III‐R DIB‐R DIPB‐R	EDs without specification, AN, BN, EDNOS	233	27 (5.9)	18–35
Miller et al. (2019)	DSM‐IV‐TR	DSM‐IV SIDP‐II	AN, BN	60	Missing data	12–18
Moriya et al. (1993)	DSM‐III‐R ICD‐9	DIB ICD‐9	EDs without specification, AN, BN	32	22.6 (3.6)	18–30
Perugi et al. (2011)	DSM‐III‐R	DSM‐IV SCID‐II	AN, BN	46	30 (7.7)	16–55
Reas et al. (2013) (Reas _ 1)	MINI	DSM‐III‐R or DSM‐IV SCID‐II	EDs without specification, AN, BN, EDNOS	737	Missing data	18–65
Reas et al. (2016) (Reas _ 2)	MINI	DSM‐IV SCID‐II	BN	282	29.9 (7.42)	18–65
Reas et al. (2015) (Reas _ 3)	MINI	DSM‐IV SCID‐II	BN	483	Missing data	18–65
Shaker et al. (2022)	BES and DSM‐5	DSM‐IV SCID‐II	BED	70	25.8 (6.34)	18–45
Shen et al. (2018)	ICD‐9	ICD‐9‐CM	EDs without specification	292	25 (missing data)	18–59
Shenoy et al. (2019)	BEDS‐7 and DSM‐5	DSM‐IV MSI‐BPD	BED	76	Missing data	Missing data
Silberschmidt et al. (2015)	DSM‐IV SCID‐I	DSM‐IV DIPD‐IV	AN, BN	770	Women: 32.1 (10.3) Men: 33.2 (10.6)	18–65
Slotema et al. (2018)	MINI	DSM‐IV‐TR	EDs without specification	84	Good outcome: 36 (11.5) Poor outcome: 38.5 (12)	Missing data
Tadić et al. (2009)	DSM‐IV M‐CIDI	DSM‐IV SCID‐II	EDs without specification, AN	159	32.9 (9.1) Women: 32.0 (9.5) Men: 34.9 (8.7)	Missing data
Wong et al. (2010)	DSM‐IV CB‐SCID‐I	DSM‐IV SCID‐II	EDs without specification, AN, BN, EDNOS	30	33.3 (11.2)	18–64
Zanarini et al. (2004) (Zanarini _ 1)	DSM‐III‐R SCID‐I/P	DSM‐III‐R DIB‐R DIPD	EDs without specification, AN, BN, EDNOS	290	27 (6.3)	Missing data
Zanarini et al. (2021) (Zanarini _ 2)	DSM‐IV SCID‐I	DSM‐IV Adult: DIB‐R DIPD‐R Adolescent: DIB‐R CI‐BPD KID‐SCID	EDs without specification, AN, BN, EDNOS	104	Missing data	Adolescents: 13–17 Adults: 18–35
Zanarini et al. (1998) (Zanarini _ 3)	DSM‐III‐R	DSM‐III‐R DIB‐R DIPD	EDs without specification, AN, BN, EDNOS	379	27.6 (6.9)	18–50
Zlotnick et al. (2002)	DSM‐IV SCID‐I	DSM‐IV SCID SIDP‐IV	EDs without specification	37	31.4 (8.9)	Missing data

Abbreviations: AN = anorexia nervosa, BED = binge eating disorder, BES = Binge Eating Scale, BEDS = binge eating disorder screener, BN = bulimia nervosa, BPD = borderline personality disorder, BPI = borderline personality inventory, CI‐BPD = Childhood Interview for DSM‐IV Borderline Personality Disorder, DIB = Diagnostic Interview for Borderline personality disorder, DIPB = The Diagnostic Interview for Personality Disorders, DSM = Diagnostic and Statistical Manual of Mental Disorders, EAT = Eating Attitudes Test, EDs = eating disorders, EDNOS = eating disorder not otherwise specified, ICD = International Classification of Diseases, IPDE = International Personality Disorders Examination, KID SCID = Structured Clinical Interview for DSM‐IV Childhood Diagnoses, M‐CIDI = Munich Composite International Diagnostic Interview, MINI = Mini International Neuropsychiatric Interview, MSI‐BPD = McLean Screening Instrument for BPD, NIMH DISC = National Institute of Mental Health Diagnostic Interview Schedule for Children, SCID‐I = Structured Clinical Interview for DSM‐IV Axis I disorders, SCID‐II = Structured Clinical Interview for DSM‐IV Axis II disorders, SIDP = Structured Interview for DSM‐IV Personality, USA = United States of America.

3.3. Risk of Bias in Studies

The global score pertaining to the methodological quality assessment of included studies varied from four to eight with a mean of 5.4 ± 1.07. This corresponds to a moderate risk. More precisely, a low risk, a moderate risk, and a high risk of bias were respectively encountered for four (11.8%), 22 (64.7%) and eight (23.5%) studies. The details concerning the risk of bias appraisal are shown in Table 4.

TABLE 4.

Risk of bias assessment according to the Joanna Briggs Institute Critical Appraisal Checklist for studies reporting prevalence data.

Study	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Total score regarding study risk of bias	Risk of bias
Banzhaf et al. (2012)	N	N	N	Y	Y	Y	Y	Y	Y	6	M
Barral et al. (2016)	N	U	U	Y	Y	Y	U	N	Y	4	H
Black et al. (2007)	N	Y	U	Y	Y	Y	Y	Y	Y	7	L
Casellas‐Pujol et al. (2024)	N	N	Y	Y	Y	Y	Y	Y	U	6	M
Chen et al. (2009)	N	N	U	Y	Y	Y	U	N	Y	4	H
Cortés‐García et al. (2021)	N	U	U	Y	Y	Y	Y	Y	N	5	M
Del Pozo et al. (2018)	N	U	U	Y	Y	Y	Y	N	Y	5	M
Dulit et al. (1994)	N	U	U	Y	Y	Y	Y	U	N	5	M
Farrés et al. (2018)	N	U	U	Y	Y	N	U	Y	Y	4	H
Frías et al. (2017)	N	U	U	Y	Y	Y	Y	N	Y	5	M
Johnson et al. (2003)	N	U	Y	Y	Y	Y	Y	Y	Y	7	L
Kahl et al. (2012)	N	U	U	Y	Y	Y	U	Y	Y	5	M
Khosravi et al. (2020)	N	Y	Y	Y	Y	Y	Y	N	Y	7	L
Kockler et al. (2017)	N	U	U	Y	Y	Y	Y	U	Y	5	M
Kot et al. (2023)	N	N	Y	Y	Y	Y	Y	Y	U	6	M
López‐Villatoro et al. (2025)	N	N	U	Y	Y	Y	Y	U	U	4	H
Marino et al. (2000)	N	U	U	Y	Y	Y	Y	Y	Y	6	M
Miller et al. (2019)	N	U	U	Y	Y	Y	U	Y	N	4	H
Moriya et al. (1993)	N	U	U	Y	Y	Y	Y	U	Y	5	M
Perugi et al. (2011)	N	U	U	Y	Y	Y	U	N	Y	4	H
Reas et al. (2013)	N	U	U	N	Y	Y	Y	Y	Y	5	M
Reas et al. (2016)	N	U	U	N	Y	Y	U	Y	Y	4	H
Reas et al. (2015)	N	U	U	Y	Y	Y	U	Y	Y	5	M
Shaker et al. (2022)	N	N	Y	Y	Y	Y	Y	N	Y	6	M
Shen et al. (2018)	Y	Y	U	Y	Y	Y	Y	Y	Y	8	L
Shenoy et al. (2019)	N	N	Y	Y	Y	Y	Y	N	Y	6	M
Silberschmidt et al. (2015)	N	U	Y	Y	Y	Y	U	Y	Y	6	M
Slotema et al. (2018)	N	U	U	Y	Y	Y	Y	Y	Y	6	M
Tadić et al. (2009)	N	U	U	Y	Y	Y	U	Y	N	4	H
Wong et al. (2010)	N	N	N	Y	Y	Y	Y	N	N	5	M
Zanarini et al. (20042004)	N	U	U	Y	Y	Y	Y	Y	Y	6	M
Zanarini et al. (2021)	N	U	U	Y	Y	Y	Y	N	Y	5	M
Zanarini et al. (1998)	N	U	U	Y	Y	Y	Y	Y	Y	6	M
Zlotnick et al. (2002)	N	U	N	Y	Y	Y	Y	U	Y	5	M

Note: List of the nine questions (Q1–Q9) applied to the studies: Q1: Was the sample frame appropriate to address the target population? Q2: Were study participants sampled in an appropriate way? Q3: Was the sample size adequate? Q4: Were the participants and the setting described in detail? Q5: Was the data analysis conducted with sufficient coverage of the identified sample? Q6: Were valid methods used for the identification of the condition? Q7: Was the condition measured in a standard, reliable way for all participants? Q8: Was there appropriate statistical analysis? Q9: Was the response rate adequate, and if no, was the low response rate managed appropriately?

Note: To calculate the total score regarding study risk of bias, the different studies' scores were summed. To attribute a score to one of the nine items (questions), consider that the answer ‘Y’ corresponds to a score of ‘1’ and the answers ‘N’ or ‘U’ or ‘NA’ correspond to a score of ‘0’ Low risk of bias (L) if ≥ 70% of questions answered ‘yes,’ moderate risk of bias (M) if ≥ 50% and < 70% of questions answered ‘yes,’ and high risk of bias (H) if < 50% of questions answered ‘yes.’

Abbreviations: N = no (high risk of bias), NA = not applicable, U = unclear (unclear risk of bias), Y = yes (low risk of bias).

3.4. Results of Syntheses

For studies that assessed EDs without specifying the type of ED, the pooled prevalence (see Figure 2) was 29.7% (95% CI = [21.6%, 38.4%]), with substantial heterogeneity (I ² = 97% with 95% CI = [96.3%, 97.7%], τ² = 0.04, p < 0.01). The H ² was 33.8 with 95% CI = [26.8%, 42.6%]. Through subgroup analyses, we found that this pooled prevalence significantly varied according to EDs evaluation tools (χ² = 339.55, with a higher frequency associated with DSM‐III‐R) and to methods of BPD identification (χ² = 319, with a higher frequency associated with DSM‐III‐R), but not according to study location, inclusion of adolescents or exclusive focus on females. Univariate random‐effects meta‐regression analyses yielded no significant results (supplementary Table S1).

FIGURE 2.

Forest plot illustrating the prevalence of ED without distinction in BPD.

With a GLMM, the proportion of EDs in studies that did not specify the type was 27.1% (95% CI = [25.1%, 29.1%]).

The pooled frequency of AN (see Figure 3) in BPD was 9.98% (95% CI = [5.7%, 15.3%]), with significant heterogeneity (I ² = 95.5% with 95% CI = [94.2%, 96.6%], τ² = 0.03, p < 0.01). The H ² was 22.3 with 95% CI = [17.4%, 29.1%]. This frequency was influenced by the tool of AN screening (χ² = 65.67, with a higher frequency for DSM‐III‐R), the one of BPD identification (χ² = 115.79, with a higher frequency for DSM‐III‐R), and the study location (χ² = 13.15). More specifically, the prevalence of AN was greater for Asia (13.3% with a 95% CI = [5.8%, 23.1%]) compared to America (9.4% with a 95% CI = [4.0%, 16.8%]) and Europe (10.6% with a 95% CI = [3.2%, 21.4%]). The prevalence amid studies that focused on females was significantly greater than other studies (24.6% vs. 7.2%, χ² = 6.56). However, the frequency of AN was not impacted by the inclusion of adolescents among participants. According to the meta‐regression analyses (supplementary Table S2), the proportion of AN was significantly correlated with the proportion of females (positive coefficient) and sample size (negative coefficient).

FIGURE 3.

Forest plot illustrating the prevalence of anorexia nervosa in BPD.

With a GLMM, the proportion of AN in BPD was 10.4% (95% CI = [6.9%, 15.3%]).

BN was encountered in 16.3% (95% CI = [12.1%, 21.1%]) of people with BPD (see Figure 4), with a noticeable heterogeneity (I ² = 93.1% with 95% CI = [90.7%, 94.9%], τ² = 0.02, p < 0.01). The H ² was 14.6 with 95% CI = [10.8%, 19.7%]. This pooled proportion significantly varied according to the study region (18.5% with 95% CI = [11.1%, 27.2%] in America, 18.9% with 95% CI = [12.7%, 25.9%] in Europe, and 5.4% with 95% CI = [2.9%, 9.5%] in Asia), the method used to explore BN (χ² = 29.41, with a higher frequency for DSM‐III‐R), as well as the tool of BPD screening (χ² = 42.55, with a higher prevalence for DSM‐III‐R). This was not the case with subgroups related to the inclusion of women or adolescents. According to the meta‐regression analyses (supplementary Table S3), the proportion of BN was significantly correlated with the proportion of females (positive coefficient) and sample size (negative coefficient).

FIGURE 4.

Forest plot illustrating the prevalence of bulimia nervosa in BPD.

With a GLMM, the proportion of BN in BPD was 15.7% (95% CI = [14.7%, 16.9%]).

The frequencies of BED (see supplementary Figure S1) and EDNOS (see supplementary Figure S2) among people with BPD were respectively 16.3% (95% CI = [6%, 30%]) and 18.8% (95% CI = [10.6%, 28.6%]), with substantial heterogeneities for both proportions. With a GLMM, these proportions were 13.5% (95% CI = [5.8%, 28.3%]) and 18.8% (95% CI = [11.2%, 30.2%]), respectively.

3.5. Reporting Biases

We did not observe any asymmetric aspects on funnel plots pertaining to the general prevalence of EDs (see Figure 5), as well as the prevalence of AN (see Figure 6) and BN (see Figure 7). This suggests the absence of publication bias, later confirmed by the Egger's test for each pooled proportion (t = −0.07 for EDs in general, t = 1.26 for AN and t = 1.22 for BN).

FIGURE 5.

Funnel plot related to the prevalence of ED in BPD.

FIGURE 6.

Funnel plot related to the prevalence of anorexia nervosa in BPD.

FIGURE 7.

Funnel plot related to the prevalence of bulimia nervosa in BPD.

3.6. Certainty of Evidence

Notwithstanding the absence of publication bias, as well as the adequacy of the source and target populations (indirectness), the certainty of our evidence was classified as low because we did not include randomised studies, considering the risk of bias of the included studies, and taking into account the substantial heterogeneities. Table 5 shows the details of the GRADE assessment of the certainty of the evidence.

TABLE 5.

Certainty of the evidence using the GRADE approach.

Components of the GRADE approach
No. reports and design of included studies	Risk of bias	Heterogeneity	Indirectness	Imprecision	Publication bias	Effect size (prevalence with 95% CI)	Certainty
Results related to EDs (without subtype specification)
20 reports, cross‐sectional design	Moderate	Very serious	No	Serious	No	29.7% (95% CI = [21.6%, 38.4%])	Low
Results related to anorexia nervosa
20 reports, cross‐sectional design	Moderate	Very serious	No	Serious	No	9.98% (95% CI = [5.6%, 15.3%])	Low
Results related to bulimia nervosa
20 reports, cross‐sectional design	Moderate	Very serious	No	Serious	No	16.3% (95% CI = [12.1%, 21.1%])	Low

4. Discussion

We conducted a systematic review and meta‐analysis to investigate the prevalence of EDs in patients with BPD. The pooled prevalence of EDs in people with BPD (according to studies that did not categorize the type of ED) was 29.7%, while the frequencies of AN, BN, BED and EDNOS were 9.98%, 16.3%, 16.3% and 18.8%, respectively. With a GLMM, these proportions were 27.1%, 10.4%, 15.7%, 13.5% and 18.8%, respectively. The GLMM analysis yielded frequencies that were usually lower than those obtained through meta‐analyses with the Freeman–Tukey transformation (Lin and Chu 2020). GLMMs are a robust alternative to double arcsine transformations in meta‐analysis of proportions, especially in cases of extreme proportions/sparse data (Lin and Xu 2020; Schwarzer et al. 2019).

The pooled frequencies resulting from the analyses were particularly influenced by the tools/criteria used to identify the ED and BPD (results for EDs in general, AN and BN, with higher frequencies associated with the use of DSM‐III‐R criteria), the study site (results for AN and BN) and the exclusive inclusion of women (results for AN). Additionally, both the proportion of women and the sample size significantly influenced the pooled frequency of AN and BN. The overall risk of bias was moderate, there was no publication bias, and the certainty levels of our evidence were low, mainly due to the significant heterogeneity of our results.

4.1. General Interpretation of the Results in the Context of Other Evidence

The prevalence of EDs in patients with BPD is significantly higher than the prevalence in the general population (Galmiche et al. 2019), and some common determinants may explain the frequent comorbidity between these two conditions. For instance, childhood traumatic experiences are usually associated with EDs and BPD, respectively. Indeed, numerous studies have reported that patients with EDs previously experienced emotional or physical neglect, emotional abuse, physical abuse, sexual abuse or peer bullying (Afifi et al. 2017; Caslini et al. 2016; Chuku et al. 2023; Pignatelli et al. 2017; Piran et al. 1988; Rabito‐Alcón et al. 2021; Rorty et al. 1994; Rorty and Yager 1996; Schmidt et al. 1993; Stuart et al. 1990). Similarly, childhood traumatic events, especially sexual abuse, have been found to play a critical role in the development, severity and prognosis of BPD (de Aquino Ferreira et al. 2018; Leichsenring et al. 2024; Yuan et al. 2023). According to various literature reports, childhood trauma and neglect can negatively affect emotion regulation, increase self‐harm through impulsivity, impair attentional skills, promote dissociative symptoms and body dissatisfaction, and alter self‐esteem (Bödicker et al. 2022; Chu and Dill 1990; Fox and Power 2009; P. Herzog et al. 2022; Racine and Wildes 2015; Richard‐Lepouriel et al. 2019; Rorty and Yager 1996; Scheffers et al. 2017; Waller et al. 1993). These negative life events can also lead to impairments in interpersonal relationships and changes in emotion recognition abilities, with an increased predisposition to alexithymia and despair (Brown et al. 2016; Chiu et al. 2019; P. Herzog et al. 2022; Hund and Espelage 2006; Maneiro et al. 2023; Rorty and Yager 1996). Most of these psychological functioning patterns are common to BPD and EDs, especially BN and BED (Crosby et al. 2009; Fox and Power 2009; P. Herzog et al. 2022; Siep et al. 2011). This may explain the higher frequency of these two subtypes of EDs in patients with BPD (Crosby et al. 2009; P. Herzog et al. 2022; Myers et al. 2006; Siep et al. 2011), and there appears to be a direct link between eating compulsions and impulsivity (Newton 2019; Newton et al. 1993). Emotion stability and mood regulation are also involved in compulsive hyperphagia, purging, restriction and self‐injurious behaviours that are common in BPD (Andover and Morris 2014; Brustenghi et al., n.d.; Casper 1983; Cooper et al. 2004; Corstorphine 2006; Crocq and Guelfi 2015; Johnson and Connors 1987; Johnson and Larson 1982; Rorty and Yager 1996; Walenda et al. 2021; Wang et al. 2018). Adverse childhood experiences may later be associated with alcohol and other SUD, and these disorders are common in BPD, as well as in EDs (Hailes et al. 2019; McKay et al. 2022; Sebalo et al. 2023). Indeed, comorbid SUD is frequent in patients with BPD and may have a detrimental impact on the course of the illness, particularly through an increased risk of self‐harm behaviours, including suicidal attempts (Crocq and Guelfi 2015; Euler et al. 2015; Kienast et al. 2014). In parallel, the pooled lifetime prevalence of any comorbid SUD in patients with EDs is reported to be over 20% in some studies, and BN, as well as BED, appears to have a significantly higher association with SUD (Bahji et al. 2019; Bogusz et al. 2021; Devoe et al. 2021; Eskander et al. 2020; Fouladi et al. 2015).

The association between BPD and EDs, particularly BED and BN, may also endorse a neuroanatomical dimension. Indeed, alterations of fronto‐limbic networks, especially in the amygdala, hippocampus, prefrontal cortex (PFC) and orbitofrontal cortex, have been reported in BPD patients (Ding and Hu 2021; Frank 2013; Reinsberg et al. 2023). Nenadić and colleagues also found orbitofrontal structural deficits in BPD (Nenadić et al. 2020). On the other hand, a weaker functional connectivity between the left lateral OFC and the right dorsolateral PFC, as well as the right precuneus, has been reported in BED and BN (Ahn et al. 2022). In AN, neuroimaging studies have reported some alterations in hypothalamic activation, as well as dysfunction in the amygdala, frontal areas and insula (Chae and Lee 2023; Holsen et al. 2012; Lawson et al. 2013). Structural and functional impairments in orbitofrontal and cingulate regions have been associated with EDs (Ahn et al. 2022; Frank 2013; Godet et al. 2022). On a biological basis, some studies have suggested the role of oxidative/biological alterations and the implication of neuroplasticity, commonly represented by the brain‐derived neurotrophic factor (BDNF), in the comorbidity between EDs and BPD (López‐Villatoro et al. 2024; Ruiz‐Guerrero et al. 2023; Thaler et al. 2014). Indeed, it has been reported that patients with BPD have lower levels of BDNF and higher levels of tumour necrosis factor (TNF) α and interleukin (IL) 6 in peripheral blood, which are associated with increased plasma levels of oxidative stress markers (Forte et al. 2023; Ruiz‐Guerrero et al. 2023). A systematic review found that women with BN‐BPD comorbidity have increased methylation of the BDNF compared to women with BN/bulimia spectrum disorder or healthy controls (McDonald 2019). They also reported higher rates of childhood sexual/physical abuse in women with BN and BPD (McDonald 2019). On the other hand, a meta‐analysis comparing 795 subjects with EDs and 552 controls found that patients with EDs had significantly lower levels of BDNF (Shobeiri et al. 2022).

Studies that included only women showed a significantly higher frequency of AN, and this finding seems in accordance with literature reports. Previous research (original articles and synthesis project) found significantly higher proportions of AN in women when compared to men (Galmiche et al. 2019; van Eeden et al. 2021). As reported by other meta‐analytic reviews on EDs prevalence in the general population, the respective frequencies of EDs (without specification), AN and BN varied significantly according to the diagnostic/screening tools, considering that the DSM‐IV was the most commonly reported diagnostic approach (Galmiche et al. 2019; Qian et al. 2022). Indeed, the prevalence variability of EDs can be explained by the changeability of diagnostic criteria, but also the interview method, especially when comparing self‐report with face‐to‐face interview (Galmiche et al. 2019; Qian et al. 2022). For example, according to the DSM criteria before 2013, BED was classified as an EDNOS and not as an individual diagnosis. This may explain the similar frequencies found in this study regarding these two types of ED (Call et al. 2013). Regarding the influence of the study location, our results concerning BN are consistent with most of the literature (Makino et al. 2004; Qian et al. 2022). For instance, a meta‐analytic review showed that the lifetime and 12‐month prevalence rates of BN were 7.3‐fold and 1.2‐fold higher, respectively, in Western countries than in Asian countries (Qian et al. 2022). In contrast, our results showing a higher frequency of AN in Asia are discrepant to some reports in the literature. Indeed, Qian and colleagues found that the lifetime and 12‐month prevalence of AN in Western countries were 21 times and 2 times higher, respectively, than in Asian countries (Qian et al. 2022). This could be explained by differences in sample size, with 172 participants included for Asia (3 studies), 1390 for the Americas (8 studies) and 1428 for Europe (seven studies). However, there appears to be an upward trend in the burden of EDs in Asian countries (Pike and Dunne 2015; Thomas et al. 2016), as shown in the systematic review by Galmiche et al. with point prevalence weighted means of 4.6% in America, 3.5% in Asia and 2.2% in Europe (Galmiche et al. 2019).

The inclusion of adolescents in the selected studies did not influence the rates of any type of ED. This may be related to the small number of studies and participants included in the analyses, but also to the fact that the issue of BPD diagnosis in adolescent populations is still a source of debate (Griffiths 2011; Laurenssen et al. 2013).

4.2. Strength and Limitations of the Review

Although this is one of the first meta‐analytic reviews to assess the prevalence of ED in people with BPD, our findings should be interpreted with caution, taking into account a number of limitations. More specifically, we have limitations related to the external validity of individual studies and limitations affecting the internal validity of selected research papers. With regard to external validity, most of the included samples concerned inpatient populations and people with outpatient care, which limits the generalisability of our findings. Other limitations include discrepancies in the proportion of females, the low representativeness of some regions of the world, such as African regions, and the low number of adolescents in the different studies. The latter is even more important considering the possibility of early diagnosis of BPD in adolescent populations (Guilé et al. 2018; Videler et al. 2019). As other limitations, we have the significant heterogeneities related to our recapitulative results, as well as the multiplicity of screening/diagnostic tools/criteria for ED and BPD. In fact, this variability could have influenced the pooled prevalence and, at least partially, the level of certainty of our evidence, which was low. Notably, there was no publication bias and the risk of bias was moderate. It should be noted that there were no specific data on several types of EDs, such as ARFID, OSFED, pica, rumination disorder and UFED.

4.3. Implications of the Results for Practice, Policy and Future Research

This systematic review draws the attention of health professionals working with people with BPD to the identification of EDs, especially as EDs may negatively influence the treatment and course of BPD. Our review could also be used to design other observational studies on the burden of EDs in people with BPD, with optimisation of methodological aspects to improve generalisability (e.g., by including more adolescent populations) and to reduce heterogeneity of results (by harmonising the tools used to identify the conditions concerned). This could also increase the certainty of evidence from further potential meta‐analyses. Observational and meta‐analytic studies on potential correlates of EDs in BPD could also be of interest to identify determinants that could be therapeutic targets.

5. Conclusions

Through this systematic review and meta‐analysis, we found a high prevalence of EDs and their subtypes in patients with BPD. Through subgroup analyses, we found some influencing factors such as screening methods for EDs and BPD identification, study location and exclusive inclusion of women (for AN). Notwithstanding the need to strengthen the data and knowledge on this topic by other studies, this meta‐analytic synthesis draws the attention of clinicians to the need to consider the comorbidity represented by EDs in people with BPD.

5.1. Public Significance

This study highlights the prevalence of EDs across the BPD population. These findings suggest the importance of assessing and monitoring all individuals with borderline personality disorder for the presence of EDs.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Univariate random‐effects meta‐regression analyses for the prevalence of eating disorders (without specification) in borderline personality disorder.

Table S2: Univariate random‐effects meta‐regression analyses for the prevalence of anorexia nervosa in borderline personality disorder.

Table S3: Univariate random‐effects meta‐regression analyses for the prevalence of bulimia nervosa in borderline personality disorder.

Figure S1: Forest plot illustrating the prevalence of BED in BPD.

Figure S2: Forest plot illustrating the prevalence of EDNOS in BPD.

Paudex, T. , Endomba F., Forestier N., Pierandrei A., Chauvet‐Gélinier J.-C., and Pinoit J.-M.. 2025. “Prevalence of Eating Disorders in People With Borderline Personality Disorder: A Systematic Review and Meta‐Analysis.” Clinical Psychology & Psychotherapy 32, no. 5: e70150. 10.1002/cpp.70150.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.