An initial literature search was conducted on the 12^th of August 2022 using the Medline, PsychINFO, and Scopus databases, which cover literature from psychological and other allied health disciplines. Search terms were (a) “Anorexia Nervosa” OR “Eating Disorder” OR “Anore*” and (b) “Mentalisation” OR “Mentali#ing” OR “Theory of Mind” OR “Mental flexib*” OR “Perspective Taking” OR “emotion* awar*” OR “emotion* perception” OR “emotion* detection” OR “emotion* identification” OR “emotion* recognition” OR “Cognitive Empathy” OR “Empathy” OR “Empathy Quotient” OR “Empath*” OR “EQ” OR “compassion*” OR “Emotion* Contagion.” Key words were selected based on the variables of interest for review, the author’s expertise, and by screening key words often presented in relevant literature. Searches in each database were limited to English language and peer-reviewed articles. This search was repeated on the 19^th of August 2023. One reviewer (IEG) completed searchers and compiled outcomes into the RAYAAN article screening platform (51). One additional paper that was not identified by the initial search strategy but fit the above specified criteria was included by the researchers. To check that any additional papers had not similarly been missed by the initial search, an expanded search was conducted in December 2023 using the following additional search terms: “Social Cognition” OR “Affective Cognition” OR “Reading the Minds in the Eyes” OR “The Films Expressions Task” OR “The Multifaceted Empathy Test” OR “The Frith-Happé Animations.” No additional papers were identified from this search.

As depicted in Figure 1 , the initial search yielded 1014 articles. 403 duplicates were removed, and two reviewers (IEG and MK) then completed an independent and blind title and abstract screening of all 638 articles. Any conflicts from this screening were resolved through mutual reassessment of the articles. 6.4% of the 638 articles presented a conflict, and, after resolution, 96 articles remained to be screened as full reports. From this pool of articles, 6 met inclusion criteria and were thus included in the present review. No additional articles were found on the repeated search in August or December, 2023. However, one additional article was found by the researchers outside of the formal search process. As such, 7 articles are included in the review.

The following details were extracted from each included article: first author, year of publication, overall and subgroup sample size, age, body mass index (BMI), biological sex percentages of the sample, and outcome measures used. All data were manually extracted from each included article.

The quality of included studies was assessed using the Newcastle-Ottawa Scale (NOS) (52), which evaluates studies in relation to three categories: participant selection (representativeness, size, non-respondents), comparison between groups in the study (ascertainment of screening, comparability of subjects), and outcome assessment (assessment, statistical test). There are four criteria for sample selection, two criteria for comparison, and three criteria for outcome assessment. The NOS uses a star-based system where each study can obtain one star for each assessed criterion in the selection and outcome criterion and a maximum of two stars for the comparability criterion, thereby achieving a maximum of nine stars overall. Studies are then determined to be either unsatisfactory (0-4 stars), satisfactory (5-6 stars), good (7-8 stars), or very good (9-10 stars) based on their star rating.

A summary of these data is provided in Table 1 . Of the seven included studies the average age range was between 18-30. Half of the studies had an all-female sample; however, all studies’ samples were above 90% female. AN diagnosis was determined using the DSM-5 in three studies (55–57), using the DSM-IV-TR in the three studies (14, 54, 58), and using the ICD-10 in one study (53). All seven articles employed quantitative data analysis approaches, but each investigates different outcome measures to assess empathy.

As seen in Table 2 , only two studies received a ‘good’ rating. All other studies were ‘satisfactory’, and one study was ‘unsatisfactory’. All studies had representative samples (employed random sampling), used validated measures to confirm AN diagnosis and selected appropriate statistical tests to analyse data. The majority of studies had sound response rates from participants with few dropouts. However, only two studies explicitly investigated and accounted for potential confounds such as gender and comorbidities, e.g. anxiety, depression and ASD, and only one study included a justification for their sample size by providing a power analysis. All studies failed to either report or utilise blinded assessment.

Two articles investigated emotional empathy (56, 58). While the results of both studies indicated superior performance on emotional empathy tasks in control participants, the small effect only reached statistical significance in one of these studies (56) Both studies also indicated that the AN remission group performed better than the AN acute group but worse than the non-clinical comparison group, though the effects were small and not significant. Additionally, both articles observed a significant positive correlation between BMI and emotional empathy, further supporting that emotional empathy improves as weight increases. Moreover, 58 also noted a significant difference between the AN acute group and the non-clinical comparison group. As both studies reported worse performance by the AN groups, this outcome confirms emotional empathy deficits in AN observed in previous literature (1–4).

One study found that the AN remission group performed significantly better than the AN acute group with a large effect size, and performed comparably to the health controls group when recognising negative emotional states (14). Additionally, there was no significant difference between groups for positive valance, but the data indicated that the AN acute group performed (non-significantly) the worst, followed by the AN remission and then health controls group, again with small effects. There was, however, a small effect size involving the AN remission group, whereby this group demonstrated superior performance to the acute AN group, and inferior performance to the healthy controls group (14). Once again, for neutral states, the same pattern of performance was observed where the AN acute group performed worst, albeit this was not statistically significant, followed by the AN remission and then the healthy controls group, which was again not statistically significant. Although not significant, four (14, 28, 55, 56) of the six articles investigating cognitive empathy observed that the healthy controls group demonstrated the highest accuracy in emotion recognition followed by AN remission group and then the AN acute group.

Moreover, one study (56) reported a small but significant positive correlation between BMI and cognitive empathy. Furthermore, one study discerned that the AN acute group performed significantly worse than the non-clinical comparison group, with a large effect size, once again confirming cognitive empathy deficits in AN in line with previous literature (14).

In addition, one study identified that both the remission and acute AN groups performed significantly worse than the non-clinical comparison group to a small degree, suggesting that even individuals with AN who have achieved weight restoration still may show impairments in theory of mind (54). Similarly, one study (53) found that the AN acute group performed better than the AN remission and HC groups on one outcome measure. However, these differences were not observed on any of the three other outcome measures. By contrast, Leslie et al. (57), did not observe any groups differences or trends in cognitive empathy outcomes.

There are several limitations within this systematic review that are important to consider when interpreting findings. Firstly, only a small number of studies were obtained, particularly for those assessing emotional empathy, and used various measures. Studies reviewed were often underpowered, which was highlighted by the number of patterns in the data that did not reach significance.The diversity in measures used for each study presents an additional confound. One study (58) gathered self-report data, requiring participants to reflect on past experiences, while all other studies utilised performance-based measures to capture empathy levels at present. This made comparing results and drawing conclusions difficult and highlights the need for additional research in this area.

Additionally, the samples within each included study overwhelmingly represented female participants. While a diagnosis of AN is more common among biological females compared to males (40) and thus sample demographics are generally aligned with clinical expectations, AN does still occur in males and there is, therefore, a gap in our understanding of the male profile of AN and the consequent neurocognitive implications. As a result, cognitive and emotional empathy could not be sufficiently examined in males. Indeed, a possible area of exploration for future research could be to examine possible differences in empathy as a factor of biological sex and also of gender to determine if there are discrepancies. Indeed, there is a higher prevalence of AN among gender diverse populations (62) and is therefore important to consider in addition to biological sex. Any differences are likely beneficial to consider in treatment to ensure intervention effectiveness as empathy deficits may impact upon the therapeutic relationship.

Finally, a factor not explored yet in the literature is the time spent in recovery. Those in remission from AN whose weight has been clinically restored may demonstrate increasing improvements in their empathy skills as they continue to recover. Further suggestions for future research, therefore, relate to the comparison of empathy in acute AN and remission across the life span, between biological sexes and among gender diverse populations, and the longitudinal recovery of AN and how empathetic ability changes across this time span. Additionally, future research should consider investigating other areas of neurocognition which may be affected be starvation such as executive functioning and memory.

While it is generally noted in the literature that people with AN present with some degree of potentially domain specific (18) reduced empathic ability, the limited literature considered in this present review provides incentive for deeper investigation of the role of starvation on neurocognitive performance, particularly in the socio-cognitive domain. However, understanding the underpinnings of why this impairment is present may be crucial in informing treatment decision making and altering ‘gold- standard’ intervention for a condition that is proven difficult to treat with long-term efficacy. The current patterns in the data are inconsistent, which could be attributable to the varying methodological and sample characteristics of each study. Additionally, studies predominantly reported non-significant outcomes. Our findings, therefore, suggest that while emotional empathy and cognitive empathy may improve throughout recovery, further investigation utilising comprehensive measures of empathy is necessary. It is anticipated that these findings will assist clinicians treating AN to account for these empathetic deficits when building a therapeutic relationship with clients from this population, setting client goals, and delivering interventions to enhance treatment outcomes.