Your body, my experience: A systematic review of embodiment illusions as a function of and method to improve body image disturbance

Multisensory integration and bodily-self-perception

Bodily self-perception results from combining real-time sensory inputs from two or more modalities, involving exteroception (i.e., visual and tactile inputs), interoception (i.e., internal body input), and proprioception (i.e., the sense of body position/movement in space; Brizzi et al., 2023). Multisensory integration enables us to continually update how we perceive and recognise ourselves (Longo et al., 2008), as evidenced via multisensory bodily (embodiment) illusions (Kilteni et al., 2015).

For instance, the classic rubber hand illusion (RHI; Figure 1) involves a three-way interaction between vision (i.e., seeing a rubber hand stroked), touch (i.e., feeling one’s [unseen] hand receive stroking), and proprioception (i.e., sensing the spatial location of one’s hand; Botvinick & Cohen, 1998). Synchronous (but typically not asynchronous) interpersonal multisensory stimulation involving both hands generally produces the temporary illusorily experience of ‘embodying’ the artificial hand (Longo et al., 2008).

• Download figure
• Open in new tab

Figure 1 Illustration of the ‘Classic’ Rubber Hand Illusion

Note. A participant’s hand is obscured from view while a rubber hand is placed in view. Both hands are simultaneously stroked. Image used with permission from Mussap & Salton (2006).

‘Full-body’ and ‘enfacement’ illusions—which employ the same multisensory principles as the RHI to one’s whole body and face, respectively—have also been developed. Full-body illusions (see Figure 2) are typically generated using immersive virtual reality (VR). Whilst wearing a head-mounted display, participants observe a virtual body (avatar) that visually substitutes their own (unseen) body from a first-person perspective (Normand et al., 2011; Piryankova et al., 2014). Visuo-tactile (i.e., touches or vibrations delivered to the participant’s and avatar’s corresponding body parts) or visuo-motor (i.e., the participant’s and avatar’s movements correspond) stimulation can induce embodiment.

• Download figure
• Open in new tab

Figure 2 Example of the Procedure Used to Induce the Full-Body Illusion in a Virtual Reality Setting

Note. Participant observes an avatar’s abdomen being stroked synchronously with their own abdomen from a first-person perspective via a head-mounted display (top right image). Tactile stimulation is administered by the experimenter (main image). Figure used with permission from Keizer et al., 2016.

Enfacement illusions typically involve watching another person’s/model’s face (e.g., via a computer screen or sitting opposite) whilst experiencing visuo-motor (i.e., participants mimic the actor’s facial expressions) or visuo-tactile (i.e., participant’s and actor’s faces are stroked with a brush or cotton tip) stimulation (e.g., Ma et al., 2016).

Embodiment illusions are typically measured subjectively via self-report questionnaires assessing ownership, agency, and self-location over the other body part, and objectively via perceptual or behavioural tasks (e.g., body size estimation). Please see Appendix A for a full summary of measures.

Embodiment illusions are generally interpreted within a predictive coding framework of brain functioning (for review see Tsakiris, 2017). This theory proposes that current self-perception relies on the brain consistently interpreting lower-level (bottom-up) sensory information under a higher-level (top-down) predictive model of what is most likely to be ‘self’ based on past perceptual experiences (for review see Apps & Tsakiris, 2013). Incongruence between the predicted and actual sensory input generates prediction errors which are reconciled by re-aligning the incoming signals with the predictive model. For example, in the RHI, the brain’s initial predictive model of the rubber hand as ‘not-self’ based on internal proprioceptive and interoceptive information (“my hand is placed under the box, not where the rubber hand is”), becomes incongruent with incoming external visual and tactile input (“I feel my hand receive stroking as I see the rubber hand receive the same”). This predictive error is reconciled by updating the model to now consider the rubber hand as one’s own (Tsakiris, 2017). Whether and how these theoretical principles shed light on distorted self-perception among individuals with BID will now be considered.

Embodiment illusions in individuals with BID

Prior research has revealed higher-order perceptual disturbances amongst individuals with clinically significant (i.e., EDs, BDD) and sub-threshold BID across visual, tactile (Engel & Keizer, 2017), proprioceptive (Eshkevari et al., 2012), and interoceptive (Badoud & Tsakiris, 2017) sensory domains. Researchers have increasingly begun to use embodiment illusions to investigate whether and how disturbances in multisensory integration influence body perception in clinical populations (for reviews see Moseley et al., 2012; Matamala-Gomez et al., 2021). To our knowledge, only one non-systematic review by Crespi and Dinsdale (2019) has examined susceptibility to embodiment illusions across a range of mental disorders and conditions: specifically, those believed to experience either heightened (e.g., EDs, BDD, psychotic disorders) or reduced susceptibility (e.g., autism spectrum disorder) to embodiment. Interestingly, five (71%) out of seven ED/BDD-based studies observed greater susceptibility among those with high (relative to low/no) symptoms. However, their review was restricted to the RHI, thus, failing to capture studies examining other embodiment paradigms. No researchers have systematically reviewed the literature regarding susceptibility to embodiment illusions as a function of BID.

Using embodiment illusions to improve BID

Although evidence-based treatments for EDs and BDD have significantly improved in quality and quantity over recent years, they often achieve modest results (for review see Ziser et al., 2018). For instance, the widely accepted cognitive behavioural therapy (CBT; Hay, 2020) typically focuses on weight restoration or minimising cognitive-affective or behavioural symptoms (e.g., fear of weight gain, loss of control eating; Kass et al., 2013). Generally, only 30-50% of ED patients report cessation of symptoms following CBT, with the remaining exhibiting partial remission, no or minimal improvement, or premature withdrawal from treatment (Atwood & Friedman, 2020). Similarly, typically only 40-50% of BDD patients respond to CBT (Öst et al., 2019), with many requiring long-term treatment (Flygare et al., 2020).

Intervention strategies targeting distorted self-perception are less common than those targeting cognitive-affective or behavioural symptoms and the efficacy of current perceptual approaches (e.g., mirror exposure therapy) remains unclear (for review see Ziser et al., 2018). Effective strategies targeting distorted body perception are required given that these distortions are a risk factor for cognitive-affective disturbance (Keizer et al., 2014) and are highly associated with ED (Engel & Keizer, 2017) and BDD (Khemlani-Patel et al., 2011) maintenance and relapse.

Embodiment illusions target the processes underlying self-perception and demonstrate how it is a dynamic and malleable construct. Their potential benefit is that the amalgamation of the embodiment model into one’s self-representation can be used to temporarily update, and thus improve BID (Keizer et al., 2016; Preston & Ehrsson, 2014). That is, one may be able to see their features more positively (e.g., torso width or some other perceived defect) due to the assimilation of the more socio-culturally desirable features of the embodiment model. Given the general tendency for individuals with BID to overestimate their body size (Gardner & Brown, 2014), reductions in body size estimation (and misestimation) could be considered an improvement.

Prior non-systematic reviews have provided a clinical rationale for the use of embodiment illusions in mental health (Moseley et al., 2012), and importantly, researchers have increasingly investigated the effects of embodiment illusions on improving BID. Earlier non-systematic reviews (Matamala-Gomez et al., 2021, Riva et al., 2021) have broadly explored and supported the potential efficacy of embodiment illusions, specifically full-body illusions, for improving BID. To date, only one systematic review and meta-analysis by Turbyne et al. (2021) has evaluated the efficacy of VR-based full-body illusions in improving BID. Of the 13 reviewed studies, nine (69%) found improvements in BID post-embodiment. However, their review was restricted to full-body illusions and ‘body image’ constructs, thus failing to capture studies assessing other embodiment paradigms and ED-specific outcomes (e.g., fear of gaining weight). Moreover, the authors conceptualised improvement according to body dissatisfaction, neglecting perceptual BID. These limitations must be addressed before concluding the efficacy of using these illusions in BID populations.

Additional influencing factors

Several possible factors warrant consideration when systematically reviewing the BID literature assessing susceptibility to embodiment illusions and post-embodiment improvements. First, although BDD and EDs (including their sub-types and symptoms) share core features, from a current clinical perspective, they exhibit distinct manifestations (APA, 2013) and should be examined independently.

Second, the effects of real-world versus VR-based embodiment illusion paradigms among BID populations remain unclear and may differ; for instance, depending on brain responsiveness to different stimuli or the customisability of stimuli enabled by the paradigm.

Third, scant research has directly compared the effect of the body part (or the number of parts) stimulated in BID populations. Differences may emerge based on its cognitive-affective salience (Frederick et al., 2016) or the density of sensory receptors reflecting that region within the brain (Corniani & Saal, 2020).

Fourth, whilst embodiment has been argued to emerge regardless of the model’s visual size (Normand et al., 2011; Preston & Ehrsson, 2014), the size of the embodied body part relative to one’s body size (i.e., smaller, larger, or same-sized) may influence improvements in BID post-embodiment. Turbyne et al.’s review (2021) reported that embodying a smaller (relative to larger) body leads to greater improvements in BID: however, conclusions remain tentative as their findings were based on VR-based full-body illusions alone and the effects of embodying same-sized bodies are unclear.

Finally, embodiment illusions are typically sensitive to the synchronous temporal aspects of visual and tactile stimulation, with asynchronous stimulation typically used as a control condition (Longo et al., 2008). However, it is unclear whether the effects of temporal synchrony during interpersonal multisensory stimulation in BID populations may manifest differently. Turbyne et al.’s (2021) review found that individuals with BID still experienced embodiment and changes in BID post-asynchronous stimulation, though to a lesser extent than post-synchronous stimulation.

The current systematic review

Despite the growing number of studies, there has been no attempt to systematically review the embodiment illusion literature concerning both susceptibility and effectiveness as a function of BID. The present systematic review has two primary aims. First, to examine whether susceptibility to embodiment illusions differs as a function of varying levels of BID. This may better elucidate whether and/or how individuals with high BID exhibit specific deficits/distortions in self-perception. Second, to assess whether embodiment illusions can improve BID. This may assist in developing alternative prevention, early intervention, and treatment approaches targeting EDs and BDD. We will also explore whether additional population-related factors (ED or BDD diagnosis) and embodiment-related phenomena (e.g., body part targeted) impact susceptibility and improvements to enhance insight into how these illusions can be most effectively conducted.

We hypothesise that (1) individuals with high (relative to low/no) BID will be more susceptible to embodiment illusions, and (2) individuals (irrespective of pre-existing BID) will show improvements in BID after experiencing embodiment illusions.

Selection of studies

The primary aims and methods of the current systematic review were pre-registered in the PROSPERO database (ID CRD42021277687). The methodology for the present review followed the Preferred Reporting for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009). Relevant studies were identified by searching the international online databases PsycINFO, MEDLINE, EMBASE, Web of Science, and CINAHL. ProQuest Dissertations and Theses Global were also searched to identify relevant unpublished (grey) literature. Citation searching of included articles was also conducted. An initial search was conducted on August 25, 2021, and then updated on May 25, 2022, and December 23, 2022, following identical search and coding procedures.

For each database, keywords (and related terms) from the embodiment illusion literature were entered into the field of keyword, title, and/or abstract. They were combined with additional terms (synonyms) related to the key concepts of (1) body image, (2) eating symptomatology, and (3) dysmorphic symptomatology, using the conjunction “AND” and separate searches for each combination. The search syntax was created by the first author J. P. and checked by a librarian. Please see Appendix B for the full search syntax.

The inclusion criteria were studies that examined and measured (a) at least one multisensory bodily illusory paradigm (e.g., the RHI) concerning (b) at least one BID-related variable. This included (i) clinical EDs or BDDs (e.g., anorexia nervosa, bulimia nervosa, muscle dysmorphia) or (ii) BID-related symptoms or obesity (as a means of representing BID; for review see Weinberger et al., 2017) within a community-based sample. Improvement in BID was considered perceptual (conceptualised as reduced body size estimation or misestimation relative to baseline) or cognitive-affective (conceptualised as an improved rating on a BID-related scale relative to baseline). Studies that assessed changes in BID were included.

Exclusion criteria included: (a) studies that examined embodiment illusions and BID exclusively among clinical populations or conditions that are not of central interest to this review—e.g., schizophrenia, depression, physical condition/injury, or neurological conditions (e.g., epilepsy); (b) studies that exclusively employed neurobiological outcome measures (e.g., functional magnetic resonance imaging [fMRI]); (c) failure to draw conclusions based upon inferential statistics involving quantitative data (i.e., significance testing based upon p < .05 or adjusted levels if multiple comparisons); (d) non-empirical studies (e.g., reviews, case studies); (e) qualitative studies; (f) and if the full-text was unavailable.

Database searches identified 5,115 potential studies. The Covidence platform (Veritas Health Innovation, 2022) aided in screening and selection. Titles and abstracts were screened against the inclusion and exclusion criteria by J. P. and independently by a second reviewer (K. P.). Full texts of all remaining articles were examined independently by J. P. and K. P. to determine their eligibility, with no discrepancies between reviewers. After removing duplicates, 1,555 studies were screened via titles and abstracts, and then 99 full-text records were assessed for eligibility: 69 were excluded as they did not meet the criteria, leaving a final total of 32 studies. Figure 4 summarises the selection process and reasons for exclusion.

• Download figure
• Open in new tab

Figure 4 PRISMA Flow Chart of Systematic Review Selection Process

Note. Adapted PRISMA 2020 flow diagram for systematic reviews which include searches of databases and registers only.

Data extraction

Extracted data for the 32 studies included: (a) bibliographical data (author names, publication year); (b) aim(s) (as relevant to the current systematic review); (c) study details (country, setting; e.g., clinical, community); (d) sample and descriptive statistics (sample size, age, body mass index [BMI], gender, clinical diagnosis); (e) stimuli and measures (as relevant to the current systematic review); (f) procedure; (g) key findings (addressing the aims of the current systematic review) based upon statistical significance and effect size; and (h) strengths and limitations. Relevant data were extracted by J. P. using Covidence and cross-checked by H. L. Please see Appendix C for the summary of results table.

Included studies had significant statistical, methodological, and measurement heterogeneity. Rather than attempting a meta-analysis, we undertook a systematic narrative synthesis of results using (i) a vote count approach (i.e., X out of Y studies) based upon inferential statistics and (ii) effect size (Higgins et al., 2019; Moher et al., 2009). Cohen’s d was chosen as the measure of effect size, with effects of 0.2, 0.5, and 0.8 deemed small, medium, and large, respectively (Cohen, 1988). If Cohen’s d was not reported and relevant data was available (e.g., means and standard deviations, t-tests, or analysis of variance), then it was calculated using the online calculator Psychometrica (Lenhard, 2016).

Quality appraisal and risk of bias assessment

Methodological quality and risk of bias were assessed using Rozenblat et al.’s (2017) revised version of Downs and Black’s (1998) original criteria, adapted for the BID-related literature. Thirteen criteria assessed whether studies (a) clearly described the aims, objectives, and hypotheses; (b) clearly outlined the main interventions and/or measures in the introduction or method; (c) specified participant characteristics; (d) clearly described interventions of interest; (e) clearly described principal confounders; (f) stated characteristics of participants lost at follow-up (if relevant); (g) clearly described the main findings; (h) included exact probability values (or confidence intervals included); (i) used a representative sample of the population examined (i.e., including clinical but not convenience samples); (j) noted any “data-dredging” explicitly; (k) used appropriate statistical tests; (l) employed valid and reliable main outcome measures; (m) conducted adequate adjustment for confounding variables; and (n) had sufficient power. There was a maximum score of 15 points¹. Using a modified grading system (O’Connor et al., 2015), scores were converted to percentages, then studies were assigned a grade of “excellent” (>86%), “good” (85%-71%), “fair” (70%-50%) or “poor” (<50%). Studies were evaluated independently by two coders (J. P., H. L.) and then cross-checked, with no concerns raised.

Study characteristics

The main characteristics and results of the 32 included studies are summarised in Appendix C. Studies were published between the years 2006 and 2022 in the English language. Thirty studies were published in peer-reviewed journals and two were from the grey literature (Lavenne-Collot et al., 2022; Wolf et al., 2021). Sample sizes ranged from six to 167, with a median of 36. The total participants numbered 1,640, of which 1,339 (81.65%) were female (382 from ED samples, 13 from a BDD sample, and 944 from community-based samples or healthy control groups), 260 (15.85%) were male (five from an ED sample, four from a BDD sample, and 251 from community-based samples or healthy control groups), and 41 (2.50%) were unidentified regarding gender due to an absence of information. Nineteen studies examined females exclusively, ten examined a mixed-gender sample, two examined males exclusively, and one was gender-unidentified. The mean age of participants ranged from 17 to 37 years, though the dominant age group was early 20s. The mean BMI (kg/m²) ranged from 15.50 to 19.80 for ED samples (reflecting the various ED subtypes) and from 18.98 to 45.33 for community-based samples (the upper range reflected two studies purposely assessing obese samples). Mean BMI was not reported for the BDD sample.

Most studies were European, being conducted in Italy (n = 8), Spain (n = 7), Germany, the United Kingdom (ns= 3), France, Sweden, and the Netherlands (ns = 2). Non-European countries included Australia (n = 4) and Taiwan (n = 1). Participants with EDs or BDD were generally drawn from voluntary ED databases or body image/ED clinics in hospital settings. Most community-based samples and healthy controls were recruited through convenience sampling (e.g., university laboratories) or via community advertisement.

Regarding samples, 13 studies examined clinical conditions: specifically, anorexia nervosa (n = 9), a mixed ED group (n = 3; i.e., anorexia nervosa, bulimia nervosa, binge eating disorder, and other specified feeding and eating disorder), and BDD (n = 1). Of these studies, 31% verified diagnoses via structured/semi-structured interviews following DSM-IV (Eshkevari et al., 2012; Keizer et al., 2016), DSM-IV-TR (Kaplan et al., 2014) or DSM-V criteria (Eshkevari et al., 2014); the remaining relied on current DSM-V diagnoses made externally (e.g., within hospitals) without interview-based verification. Nine of these 13 studies utilised healthy control samples, however, only 38% of these samples were screened for BID using validated measures (e.g., Eshkevari et al., 2012; 2014). The remaining 19 studies examined a community-based population: two compared individuals with obesity (i.e., likely sub-threshold BID) to healthy-weight controls, though neither screened for BID in the latter (Scarpina et al., 2019; Tagini et al., 2020). Research designs were predominately quasi-experimental, with three studies having a longitudinal design.

Synthesis of main findings

Quality appraisal and risk of bias assessment

As Table 2 shows, the quality evaluation found that most studies were deemed fair or good (46.88% versus 40.62%), fewer were poor (12.50%), and none were excellent. The most commonly met criteria were the use of appropriate test statistics (100% of studies), exact probability values (84.38% of studies), and clear descriptions of results (81.25% of studies) and interventions of interest (78.13% of studies). Potential sources of bias included low statistical power (or failure to conduct a power analysis; 90.63%), non-representative samples (81.25% of studies), poor control of confounding variables (75.00% of studies), and non-psychometrically validated and reliable main outcome measures (53.13% of studies).

Susceptibility to embodiment illusions as a function of BID

Researchers have increasingly investigated susceptibility to embodiment illusions as a function of BID. However, only one prior review (Crespi & Dinsdale, 2019) has examined susceptibility to embodiment illusions across a range of mental disorders and conditions: observing higher susceptibility in 71% of the seven ED/BDD-based studies. This prior review was non-systematic and focused solely on the RHI. The current review, drawing from an extended body of research involving broader embodiment illusion paradigms, similarly revealed that 64% of studies reviewed reported significantly greater susceptibility among those with high (relative to low/no) BID. The effects were generally medium to large-sized, suggesting that the current findings have practical significance.

These findings align with the notion that higher levels of BID may be related to more malleable perceptual body representation, possibly due to fundamental deficits involving multisensory integration: namely, external visual input receiving more weight than either tactile and/or proprioceptive input (e.g., Eshkevari et al., 2012; Gaudio et al., 2014). Deficits in interoceptive processing have also been positively linked to embodiment susceptibility (Bekrater-Bodmann et al., 2020) and consistently observed in BID populations (for a review see Jenkinson et al., 2020). According to predictive coding accounts of EDs (Barca & Pezzulo, 2020), BID-related disorders may involve bodily-self-perception that is less dependent on incoming sensory input and more reliant on higher-level predictions (based on prior beliefs). Importantly, this interpretation may offer valuable insight into the aetiology of these disorders because prior beliefs are not just important for perceptual inference but also action selection and control (goal-directed behaviour). For instance, when considering EDs, reliance on prior beliefs about which body size is to be pursued (e.g., a very thin body; Barca & Pezzulo, 2020) over internal bodily input (e.g., hunger cues) may contribute to the maintenance of the disorder by increasing maladaptive affect and cognitions (e.g., prioritising visual body-related feedback over hunger cues). This may also pose challenges to forming and maintaining an accurate bodily representation (Barca & Pezzulo, 2020).

This account could be extended to Allocentric Lock Theory (Riva & Gaudio, 2018). This neuroscientific framework posits that individuals with EDs hold a ‘locked’ or unchanging distorted body image because they are unable to update their ‘remembered’ body (i.e., an allocentric representation that they constructed long ago) with perceptual information from their ‘experienced’ body (i.e., an egocentric representation). The minority of inconsistent findings (i.e., null between-group differences or reduced susceptibility in the high BID group) may be attributable to methodological issues, as outlined below.

Experiencing embodiment illusions as a means to improve BID

Only one prior review which examined 13 VR-based studies involving full-body illusions has considered whether embodiment illusions can impact BID (Turbyne et al., 2021); improvements were reported in 69% of studies. Using an extended number of studies (n = 24) involving both VR and real-world approaches, we revealed that 83% of studies observed significant, generally medium to large-sized, improvements in BID post-embodiment; supporting the practical and clinical utility of embodiment illusions when improving BID. Current findings, and those of others (e.g., Keizer et al., 2014; Turbyne et al., 2021), are compatible with the idea of utilising fluctuating perceptual and cognitive-affective body representations (as a result of disturbances in multisensory integration) in individuals with BID (Espeset et al., 2012); specifically, BID appears to be able to improve via the integration of an external body(part) into one’s mental bodily-representation.

Additional factors impacting susceptibility and improvement

Limitations and strengths of the included studies

Confidence in the validity and reliability of our findings should be considered against the limitations and strengths of the included studies. The qualitative assessment revealed that most studies were deemed fair or good, none were excellent, and a small portion was poor. Most studies described the embodiment procedure with a replicable level of detail, clearly reported their results, and randomised their stimulation conditions to reduce the potential confound of order effects. These practices are recommended in future research. Although consistent findings were observed across lower and higher-quality studies (e.g., Eshkevari et al., 2014; Mussap & Salton, 2006), several unaddressed limitations warrant attention.

First, measures of embodiment were substantially heterogeneous, lacked psychometric data, and were generally narrow in scope. Over 40% of studies failed to measure (e.g., Mussap & Salton, 2006) or independently assess (Metral et al., 2017) both subjective and objective embodiment. This is limiting as they offer contrasting and complementary sources of information regarding the mechanisms underlying self-perception (for review see Braun et al., 2018). For subjective embodiment, most researchers created modified versions of two commonly used RHI questionnaires (Botvinick & Cohen, 1998; Longo et al., 2008), which were not validated psychometrically and varied in the number of items, number of sub-scales (ownership, agency, and/or self-location), and scoring (totals, individual sub-scales, or individual items). Body-part size estimation, among the most frequently used measures of objective embodiment (e.g., Keizer et al., 2016), also varied considerably, including width estimates using a laser beam (e.g., Serino et al., 2020) or circumference estimates using string (e.g., Serino et al., 2016). Moreover, these typically failed to capture structural aspects (e.g., the shape of the abdomen), and again, lacked psychometric data. Increased efforts are needed to utilise valid, reliable, and consistent measures of subjective embodiment, such as Romano et al.’s (2021) RHI questionnaire or Gonzalez-Franco and Peck’s (2018) proposed (though not yet validated) standardised embodiment questionnaire. More advanced and controlled body-part measures also exist, such as the Body Image Assessment Software (Ferrer-García & Gutiérrez-Maldonado, 2008) and VR-based methods whereby participants adjust their body size in VR (Provenzano et al., 2020). Although apparent contradictions in our findings may be attributable to the measures used, the heterogeneity and lack of psychometric data inhibit conclusions.

Considerable heterogeneity regarding the scope and psychometric data of measures of BID used in studies may explain some inconsistent findings. For instance, the only study (Provenzano et al., 2019) where embodiment did not affect BID used a non-validated measure of cognitive-affective BID, whilst most studies that observed an effect employed validated measures of cognitive-affective BID or both perceptual and cognitive-affective measures. Future research should utilise a more standardised and advanced battery of valid and reliable measures that span both perceptual and cognitive-affective dimensions of BID as these may map onto objective and subjective measures, respectively.

Second, the inadequate control of confounds may explain inconsistent findings regarding susceptibility. Almost all studies that found non-significant between-group differences or observed greater susceptibility in healthy controls than the high BID groups failed to screen for pre-existing BID among the healthy controls (e.g., Kaplan et al., 2014); the remaining study observed initial body size overestimations in the healthy controls (Carey & Preston, 2019). Thus, high BID may have existed within each healthy control group, reducing the ability to detect between-group differences. Comparatively, all studies that supported the hypothesis explicitly screened for BID-related psychopathology in both groups.

Finally, most participants were adult women from European countries, which is limiting given that significant differences in BID psychopathology exist between men and women (e.g., more muscularity-orientated presentations in men; Murray et al., 2018) and adolescents and adults (e.g., difficulty eliciting change among adults; Porras-Garcia et al., 2021). Non-European Union countries (e.g., the United Kingdom and the United States) and Asia were also neglected which may limit the generalisability of findings given geographical differences involving BID (Frederick et al., 2016). Future researchers should investigate whether and how demographic differences manifest in the current research questions.

Limitations and strengths of the current review

As discussed, the most notable limitation of the current review was considerable heterogeneity across studies in the assessed populations (e.g., mixed ED group, anorexia nervosa), interventions (e.g., real-world versus VR, body part[s] targeted), and outcome measures. This limited our ability to draw comparisons and conclusions involving a meta-analysis, and/or to construct valid funnel plots indicating potential publication biases: impairing attempts to fully understand the findings including the potential intersectionality between various factors raised.

Second, the current review omitted psychiatric disorders and developmental conditions that show high comorbidity with EDs and BDD, including anxiety and depression (Keski-Rahkonen & Mustelin, 2016), and autism spectrum disorder (Huke et al., 2013). Crespi and Dinsdale’s (2019) review indicated that, relative to healthy controls, individuals with EDs and autism spectrum disorder show increased and reduced susceptibility, respectively. Hence if individuals in the current review had autism spectrum disorder-related ailments, this may underpin some of our observed contradictory findings. Future reviewers should examine studies that have assessed comorbid disorders and conditions in BDD/ED samples (e.g., Kaplan et al., 2014).

Finally, we conceptualised improvement in perceptual BID as (i) reduced body size misestimation relative to baseline (e.g. Keizer et al., 2014), or (ii) reduced body size estimation relative to baseline when data on baseline misestimation was not available (e.g., Wolf et al., 2021). The latter may be problematic as without baseline data, it is speculative that the assessed sample overestimated their body size at baseline (thus, reductions might have reflected BID increases). However, given extensive research demonstrating that BID populations (Hosseini & Padhy, 2019) and those of low and healthy weight (the majority of the included samples; Cornelissen et al., 2012) overestimate their body size, we deemed it valid to infer that the included samples generally overestimated. Nonetheless, future researchers should be more rigorous and assess misestimation.

Notwithstanding these limitations, the current review had several strengths. First, including a large number of studies across various embodiment approaches allowed a more exhaustive assessment. Second, including unpublished studies reduced the potential impact of publication bias if only peer-reviewed publications were included (Borenstein et al., 2021). Finally, our quality appraisal and risk of bias assessment ensured that studies of lower quality were interpreted tentatively, and allowed us to provide specific methodological recommendations for future researchers.

Conclusions, implications, and directions for future research

The heterogeneity and limitations in methodology across studies, as well as the lack of theoretical underpinnings in the field, ultimately warrant caution when making scientific advances on the major components of BID. Nonetheless, multiple take-home messages emerged from our review. First, insights from embodiment illusion experiments may enhance our understanding of the multisensory integration basis of perceptual distortions within BID populations, and ultimately, enhance interventions. Indeed, greater susceptibility was consistently observed in high BID populations and may indicate a more malleable bodily representation due to underlying disturbances in multisensory integration and increased weighting of higher-order predictions over lower-level sensory inputs.

Second, embodiment illusions also provided a means to manipulate and improve the malleable body representation of individuals with BID. This warrants continued attempts to incorporate embodiment illusions within clinical settings as a complementary intervention to best-practice treatments (e.g., CBT) to target more holistic dimensions of BID, and in prevention and early intervention efforts targeted towards BID-related disorders. Given the aforementioned issues associated with CBT (Atwood & Friedman, 2020), focusing on perceptual aspects via embodiment may provide a means to increase the effectiveness and completion of these current interventions.

Third, new therapeutic interventions targeting BID may also utilise insights gained from research on susceptibility to embodiment illusions. For instance, sensory and multisensory training/therapies (e.g., physical therapy, mindfulness-based interventions; Weng et al., 2021) and wearable technologies (e.g., brain/body stimulation devices; Riva et al., 2017) could increase tactile, proprioceptive, and interoceptive awareness to restore the balance between visual and tactile/internal bodily input during multisensory integration (Eshkevari et al., 2014). Ultimately, this may help to form and maintain a more realistic self-representation and thus, improve BID.

Fourth, our exploration of the various influencing factors revealed that susceptibility consistently varied as a function of BID, especially when assessing anorexia nervosa samples, using a real-world embodiment paradigm, and the hand or face to induce the illusion, irrespective of stimulation synchrony. Additionally, improvements in BID were greater when utilising VR-based embodiment illusions, body parts of high cognitive-affective salience, a smaller-sized model/avatar, and synchronous stimulation. These findings may inform future attempts in both contexts to improve the efficacy of future research, prevention, and intervention approaches in BID populations.

Additional future directions warrant attention. First, only three included studies employed longitudinal designs (e.g., Porras-Garcia et al., 2021). Whilst all found enduring improvements in BID post-embodiment, there was considerable heterogeneity in the assessed periods (ranging from a few hours to 3 months). Further research into the longevity of any improvement is required before these methods can be considered in long-term strategies targeting BID. Second, there is a clear gap in embodiment illusion research examining BDD. One study assessed susceptibility in a BDD sample, no studies have examined BDD subtypes (e.g., muscle dysmorphia), and measures of improvement typically neglect BDD-specific constructs. This research would improve understanding of distinct or shared aetiologies across ED and BDD sub-types and thus, the efficacy of embodiment illusion-based interventions. Third, researchers should consider exploring neurobiological underpinnings of susceptibility to embodiment and post-embodiment improvements in BID via neuroscience technologies (e.g., fMRI), to enrich our understanding of whether and how the brain is involved. Current research has implicated—but not directly tested—(i) the insula in susceptibility to the RHI (e.g., Crespi & Dinsdale, 2019) and (ii) interactions between the insula, cingulate cortex, and parietal cortex in the effects of illusory obesity on body dissatisfaction (Preston & Ehrsson, 2016). Finally, there is a risk that ‘passively’ generated change (i.e., independent of the subject’s will)—as arguably observed in embodiment illusions—could produce mostly immediate but then ineffective effects in terms of care and treatment. Future researchers should evaluate whether and by which means embodiment illusions are efficacious in bringing about constant change in an individual who, based on an active process of taking a stand in their symptoms, has found no better way of adaption than to develop their own disorder. Addressing such avenues for future exploration would help to facilitate the successful use of embodiment illusions among researchers and practitioners working with BID populations.