This systematic review and meta-analysis was registered with the PROSPERO database of systematic reviews (CRD42020202044). This review followed both the meta-analysis of observational studies in epidemiology (MOOSE) recommendations (37) and the preferred reporting for systematic reviews and meta-analyses (PRISMA) guidelines (38, 39).

A comprehensive electronic database search of the peer-reviewed literature was conducted in the following online databases: MEDLINE, PsycINFO, Embase, and CINAHL from their inception to May 2021. The key words included two concepts: (1) various EDs [i.e., anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), or eating disorders not otherwise specified (EDNOS)] and (2) ICDs/BA (i.e., impulse control disorder total, pathological/compulsive buying, kleptomania, trichotillomania, intermittent explosive disorder, pyromania, pathological/compulsive gambling, technology addictions, and stealing/shoplifting). The electronic database searches and exhaustive list of keywords are provided in the Supplementary Material. Two blinded reviewers (A.A. and A.P.) independently performed title, abstract and full-text article screening using the online Covidence systematic review software (40). The reference lists of included articles were hand-searched for other relevant studies that might meet the inclusion criteria and were not found through online database searching.

Peer-reviewed studies were selected by two reviewers (A.A. and A.P.) for inclusion in this review if they met the following criteria: (1) research including participants with any EDs, anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), or eating disorders not otherwise specified (EDNOS) or reported total EDs, based on any DSM/ICD criteria; (2) reported on ICDs or BAs either clinician reported or self-reported (i.e., impulse control disorder total, pathological/compulsive buying, kleptomania, trichotillomania, intermittent explosive disorder, pyromania, pathological/compulsive gambling, technology addictions, and stealing/shoplifting); (3) reported the prevalence of ICDs or BAs in EDs, (4) contained cross-sectional data for the meta-analysis. In addition, this review excluded studies that: (1) looked at the relationship between EDs and substance abuse disorders; (2) had ineligible study designs such as case reports, review articles, opinion pieces, and editorials; and (3) had insufficient data for the meta-analysis. Disagreements were first discussed in a consensus meeting, and DD made the final decision for inclusion or exclusion.

Data abstraction was completed in duplicate (AA and AS), including the following study characteristics: author, year of publication, country, the control group, total sample size, types of ICDs or BAs, types of EDs, total ED sample size, age (mean ± SD), number of females/percent female, cohort or comparison group description, and outcomes reported. For the meta-analysis, the following data were extracted: (1) author, (2) year of publication, (3) ICDs or BAs type, (4) ED type, (5) numerator representing ICDs or BAs, and (6) the denominator representing EDs or ED subtype.

Studies included in this systematic review and meta-analysis were evaluated for quality using a modified Downs and Black instrument (41) by one author (TL). The modified Downs and Black checklist for cross-sectional studies utilizes 14-items to evaluate the risk of bias and provides a total score out of 15 points for each study. Thus, higher scores indicate greater quality. In addition, this quality assessment checklist contains yes/no questions, which are applied to a series of questions regarding the quality of each study.

Due to the expected heterogeneity between studies, the estimates were initially stratified by ED subgroup, and a series of DerSimonian and Laird (42) random-effects meta-analyses were performed on eligible studies to estimate pooled proportions and 95% CIs for ICDs and BAs for all EDs within those strata. All meta-analyses in this review utilized the Freeman-Tukey double arcsine transformation, which computes the weighted pooled estimate using a variance-stabilizing transformation and then performs a back-transformation on the pooled estimate. This approach is favorable in situations where there is zero count data, as it prevents zero count studies from being dropped from the analyses creating a bias in prevalence estimates. The exact confidence interval method was used as this is considered more conservative. Statistical heterogeneity was calculated utilizing the I² statistic for subgroup meta-analysis that included four or more studies. All analyses were performed in STATA v.17 (43). To avoid double-counting in the meta-analysis, the data from the largest sample (i.e., denominator) was utilized in the case of two or more studies reporting in the same sample. Forest plots were created showing the proportion of those with either ICDs or BAs in individuals with EDs. Subgroup meta-analyses were used to distinguish the prevalence of ICDs or BAs in EDs types.

Electronic database searches identified 3,310 records; after duplicate references were removed, a total of 2,308 abstracts and titles were screened. The level of agreement between the two reviewers for screening titles and abstracts was moderate (κ = 0.60). After resolution of inconsistencies, a total of 60 studies were retrieved and reviewed in full text. Overall, 35 studies met the inclusion criteria and were included in the meta-analysis, see Figure 1.

All studies included in this review are described in detail in Tables 1, 2. Studies were published between 1980 and 2019. Most studies were conducted in North America (n = 14), followed by Europe (n = 13) and Asia (n = 8). Twenty-two studies recruited ED individuals from a hospital setting or a specific eating disorder program, nine were recruited from outpatient clinics, and four were recruited from research study cohorts.

A total of 9,646 individuals were identified as having an ED, ranging from sample sizes of 10 to 2,436 participants with EDs in individual studies. The mean age of individuals with EDs was 25.5 years (range 16.7–43.9), and the percentage of females was 97.7%.

All studies included in this systematic review were evaluated with the modified Downs and Black instrument (Supplementary Table 1). For cross-sectional studies, the average Downs and Black score was 10.4/15, indicating predominantly moderate-quality studies. Although most studies were of moderate quality, several failed to account for the effects of important covariates such as age and gender.

In studies that looked at total ICDs/BAs, random pooled estimates demonstrated a 22% comorbid prevalence in those with EDs (3 studies, N = 1,026), see Figure 2. When examining individual types, pathological/compulsive buying had the highest prevalence at 19% in those with EDs (9 studies, N = 3,338), followed by a 18% prevalence for kleptomania (9 studies, N = 2,997), 4% prevalence for intermittent explosive disorder (3 studies, N = 1,026), 3% prevalence for trichotillomania (5 studies, N = 4,890), and a 2% prevalence for pathological/compulsive gambling (5 studies, N = 2,618). Pyromania did not appear to be prevalent in those with EDs (3 studies, N = 704).

In addition, there was a 12% prevalence for compulsive internet use, although this estimate was reported only on a single study and was imprecise (1 study, N = 60). Lastly, random pooled estimates demonstrated that stealing had a prevalence of 30% in those with EDs (13 studies, N = 3,749; see Supplementary Figure 1).

In studies that looked at total ICDs/BAs by ED subtype, there was a 24% prevalence in those with BN (1 study, N = 227), see Supplementary Figure 2. When examining individual ICDs/BAs by ED subtypes, compulsive buying had the highest prevalence at 42% in those with BN (3 studies, N = 341) followed by a 24% prevalence in those with BED (1 study, N = 84), an 11% prevalence in those with AN (1 study, N = 412). Kleptomania had the highest prevalence at 62% among those with AN (1 study, N = 13), followed by 11% among those with BN (3 studies, N = 337).

For individuals with BN, there was a 13% prevalence for intermittent explosive disorder (95% CI = 0.09–0.18; 1 study, N = 227), and a 3% prevalence for trichotillomania (2 studies, N = 292). For BED individuals, there was an 18% prevalence for pathological/compulsive gambling (1 study, N = 94). Neither pathological/compulsive gambling nor pyromania appeared to be prevalent in AN and BN studies.

The prevalence of stealing was 5% in AN-R (2 studies, N = 160), 19% for AN (4 studies, N = 630), 31% for AN-BP [anorexia nervosa binge eating/purging type (2 studies, N = 133), and 45% BN (4 studies, N = 658)], see Supplementary Figure 3.

This systematic review and meta-analysis have contributed to the knowledge base by being the first review to examine the pooled prevalence of comorbid ICDs or BAs in individuals with EDs. Overall, it was estimated that nearly one-quarter (22%) of all people with EDs would experience an ICD at some point during their lives. Most ICDs were more prevalent among individuals with the binge/purge subtype of EDs, such as BN and AN-BP, relative to restrictive EDs, such as AN-R. In addition, roughly one-third (30%) of individuals with ED experienced stealing/shoplifting behaviors. The quality of the literature was rated as being moderate-quality across studies; however, several failed to report and adjust for potential confounders. Overall, we conclude that ICDs are common co-occurring conditions among individuals with EDs, and clinicians should be aware of the frequency of ICDs when providing care for people with EDs.

The prevalence of comorbid ICDs/BAs, 22%, identified in this study closely resembles the prevalence of comorbid substance use disorders (SUDs) identified by a previous meta-analysis. The pooled prevalence was 22% (71). Clinicians must be aware of ICDs since they are associated with worse ED outcomes, particularly if they are undertreated (12, 52). In a similar vein, some personality traits (e.g., perfectionism, impulsivity, cognitive rigidity, and harm avoidance), psychiatric comorbidity, age of onset (e.g., early age of onset may contribute to poor prognosis), concurrent substance use disorders have been proposed as other risk factors that may play a role in deteriorating ED prognoses (72). Most notably, our meta-analysis identified an association between ICD and BN and AN-BP subtype compared to other ED diagnoses and ED behaviors. This supports previous literature indicating that individuals with binge-purge EDs (i.e., BN and AN-B) are more likely to struggle with ICDs, linked to more significant emotion regulation and impulse control difficulties (48, 73, 74).

Although our study did not identify and examine the differences in the comorbid ICDs/BAs ascertainment method, this may have contributed to statistical heterogeneity demonstrated between studies included in this review. For example, the DSM-III, DSM-IV, DSM-5, and the ICD have different specific criteria for other ICDs/Bas (6, 75–78). Furthermore, there are several self-report ED diagnostic instruments, such as the Eating Disorder Inventory (EDI) (79), and semi-structured interviews, such as the Eating Disorder Examination (EDE) (80). However, the reliability of these tools for detecting comorbid ICD/BA comorbidity may differ between studies using self-reported checklists (e.g., Eating Disorder Examination). In addition, individuals with ED often underestimate their impairments compared to informant reports (e.g., the information supplied by persons who are familiar with the patient, such as a supervisor, friend, partner, or family member), resulting in biased ICD/BA prevalence estimates (81, 82). Therefore, the diagnostic reliability could be improved further by incorporating collateral from multiple informants, especially when working with adolescents and young adults (83, 84). However, screening instruments will have a much better positive predictive value when the base rate, or prevalence of the condition, is higher in the population being screened. To that end, as this review provides an estimate of the prevalence of comorbid ICD/BAs, it can inform the best available base rate data for these disorders to date, which can support future epidemiological investigations.

Several studies have attempted to determine psychometrically robust tools for diagnosing comorbid ICDs, but there has generally been a poor correlation between rating scales (85–87). In addition, individuals may conceal the existence of ICDs/BAs due to shame and stigma, adding a layer of complexity to the screening process (88). For example, research has shown that individuals with gambling disorders experience stigma and are often stereotyped as selfish, greedy, or irresponsible, which may cause these individuals to delay help-seeking due to shame (89). Similarly, stigma research in individuals with AN has indicated that they believe the public trivializes their experiences by viewing their behaviors as within their control and by accrediting eating disorder behaviors solely to socio-cultural factors, which delays disclosing their illness (90). It may also be necessary to screen for ICDs and BAs in all EDs rather than EDs with bulimic-type psychopathology alone, as they appear to be comorbid with restrictive AN as well (e.g., pathological buying and kleptomania). As stratification by ED subtypes did not reconcile the heterogeneity in prevalence estimates, this suggests that the different instruments used to measure BAs/ICDs may differ (91–93).

One prospect for implementing and identifying treatment for individuals with these comorbid conditions is understanding the underlying pathophysiology of ICDs/BAs and the co-occurrence of SUDs. For example, both ICDs and BAs are postulated to share similar personality and neurobiological mechanisms (94). For instance, ICDs/BAs may share higher personality traits of impulsivity and sensation-seeking, with lower measures of harm avoidance (15). Furthermore, among women with co-occurring bulimia nervosa and ICDs, lower self-directedness, higher harm avoidance, and cooperativeness appear to be shared across disorders (12). In addition, research suggests that those with EDs and ICDs/BAs share a similar propensity for impulsivity (33, 95) and may have similar deficits in brain reward circuitry (34). Adding to this, it is worth noting that SUDs frequently co-occur with ICDs/BAS with the percentage of comorbid SUDs varying between 15 and 76% depending on the type of ICD/BA (96), and all conditions share similar characteristics of impulsivity. Thus, a transdiagnostic treatment approach targeting impulsivity in these comorbid conditions may help alleviate the burden that these individuals encounter, as there is mounting evidence supporting both pharmacological and behavioral treatments for impulsivity in SUDs (97), which has the potential to translate to efficacious therapies for individuals with EDs and comorbid ICDs/BAs. Finally, another potential option for treating ICDs/BAs in those with EDs is to take a personalized and flexible approach when treating those with EDs who have high levels of impulsivity (98).

There are several strengths of this study. First, to our knowledge, this is the largest systematic review of ICDs among people with EDs and the first meta-analysis. Second, the quality of the majority of studies evaluated was fair. Third, the study was methodologically rigorous, comprehensive, and adhered to the MOOSE guidelines.

However, some limitations should be considered in the appraisal of the evidence presented by this review. One of the most considerable limitations is the limited number of studies available. Consequently, some prevalence estimates only involved one or two studies. Another significant limitation is the high heterogeneity observed across studies for most estimates. There are many potential sources to this heterogeneity, including the wide range of tools used to define ICDs and EDs, the range in study years included in the review, and different classifications of ICDs/BAs. However, a meta-regression was not performed to examine these potential sources of heterogeneity due to the limited number of studies included in the subgroup meta-analyses. While our study identified comorbid ICD diagnoses rather than symptoms, these estimates represent the presence of a clinically significant psychiatric comorbidity, which may be of more relevance to most clinicians. However, ICD/BA symptoms in the absence of a clinical diagnosis can still cause significant functional impairment, primarily if the individual endorses one symptom less than is required for a diagnosis. Furthermore, there may be differential associations between specific ICD/BAs symptoms and ED diagnoses/subtypes undetected in individual studies yet may be detected in a combined sample. As is demonstrated in prior studies and replicated in the current study, the ED symptoms of binge eating and/purging are essential in this association—and perhaps a similar finding exists at the ICD symptom level (74).

While sex proportions were reported in overall samples, sex-specific prevalence estimates were only reported by a subset of studies, precluding extensive sex-based analyses. Due to a gross underrepresentation of males in existing ED literature (99), few identified studies exploring ICD/BA comorbidity in males with ED. Hence, our study may have been underpowered to find sex differences. One meta-analysis of community studies reported the 12-month and lifetime prevalence of EDs as 2.6- and 4.2-fold higher among women than men (100). Conversely, most ICD/BA diagnoses (i.e., pyromania, gambling disorder, intermittent explosive disorder, and oppositional defiant disorder) occur more frequently in men, except for kleptomania, which occurs three times more frequently among women (101, 102). Accordingly, while we found no relevant prevalence estimates of gambling disorder in BN, an ED predominantly diagnosed in women, studies reported an 18% prevalence in BED, an ED characterized by a more balanced woman-to-men ratio. Studies examining gender differences in compulsive buying have yielded mixed results. In a German community sample (103), compulsive buying occurred equally between men and women. However, women in a Spanish community sample showed a higher propensity for compulsive buying (104). Additionally, while upwards of 80% of compulsive buyers seeking treatment are women, this may reflect the notion that women are more likely to recognize and seek help for the problem rather than true differences in prevalence between genders (105). These findings might explain the differences in prevalence for the different types of ICDs/BAs in the current study samples of primarily women (97.8%) with EDs. Another aspect that may explain the differences in prevalence might be that the incidence of ICDs/BAs also tends to vary with age. For example, while oppositional defiant disorder and conduct disorder have the most significant incidence before adolescence (101), intermittent explosive disorder occurs at any point under age 40 (106).

To that end, as this is a study-level meta-analysis, a limitation of the methods is that individual-level characteristics were not explored. There was also a limited representation of studies from all geographic regions, which limited our ability to estimate the prevalence of comorbid ICD/BAs across all continents. There was little information about ICDs/BAs in specific EDs, such as night-eating syndrome, ARFID, or atypical anorexia. Consequently, the overall prevalence estimates presented in the current study are not representative of all individuals with ED, but rather AN, BN, and BED primarily. Finally, as with all meta-analyses, we were limited by the quality and quantity of existent studies, and our results reflect only what is available in terms of existing literature.

Future research is needed to understand better the prevalence of ICD/BAs in subgroups of ED individuals, especially males, children and adolescents, and older adults. Adolescents and young adults are especially important groups regarding ICD/BAs as this age group overlaps with a neurobiologically sensitive period for developing impulse control. This characteristic appears to influence the course of these disorders. Finally, given the bidirectional relationship between ICDs/BAs and EDs, and related conditions (55, 74, 107–110)—such as substance use disorders (SUDs) and personality disorders (PD), future research should examine the corresponding prevalence of these comorbidities in alternative primary populations (e.g., the prevalence of EDs in people with ICDs/BAs). In addition, future studies could examine the bidirectional causation by conducting longitudinal studies that examine EDs and ICDs/BAs. Future research directions might also involve additional studies exploring treatment options for individuals with co-occurring EDs and ICDs/BAs, particularly pharmacological interventions, rTMS, or combining the two with or without concomitant psychotherapies.