Anorexia nervosa (AN) is a severe and life-threatening eating disorder characterized by extreme food restriction, an intense fear of gaining weight, and distorted body image (1). AN predominantly affects individuals assigned female at birth, and it is considered one of the most gender-biased mental health disorders, although recent evidence suggests that prevalence in males may be underestimated due to underdiagnosis and diagnostic bias (2, 3). Understanding this gender disparity could provide valuable insights into how biological, psychological, and sociocultural factors interact in the development of psychiatric conditions. For instance, examining AN in the context of gender-related epidemiology may offer parallels with other conditions, such as autism spectrum disorder, where diagnostic biases and gender differences also play a significant role (4–6).

However, most research has focused almost exclusively on affected females, leaving the experiences and psychological characteristics of male family members largely unexplored. Investigating the impact of AN on male family members, such as brothers, could shed light on shared genetic and environmental risk factors, contributing to a more comprehensive understanding of the disorder (7). Given that brothers are less likely to develop AN themselves, but may still share predisposing factors, they offer a unique lens through which to examine whether certain traits associated with the disorder might exist independently of its clinical expression. In this sense, unaffected male siblings represent an ideal population for identifying trait-based markers that may signal familial liability, even in the absence of overt psychopathology.

Research has identified psychological traits such as empathy, theory of mind (ToM), and alexithymia as relevant to the understanding of anorexia nervosa beyond core eating symptoms, particularly due to their impact on interpersonal functioning, emotional regulation, and social cognition (8–10). Empathy refers to the ability to understand and share the emotions of others (11), while ToM involves the capacity to infer others’ mental states (12). Alexithymia, characterized by difficulty in identifying and describing emotions, has been frequently associated with AN, contributing to emotional dysregulation and impaired interpersonal relationships (13, 14). Gender differences in these traits are well-documented in the general population, with women generally exhibiting higher levels of empathy and TOM, while men are more prone to alexithymia (15, 16). This makes it essential to examine whether male siblings of individuals with AN, who are both genetically and environmentally linked yet phenotypically different, may show subclinical levels of these traits or express them differently. These differences raise important questions about how male and female siblings of individuals with AN might experience and express these traits.

While individuals with AN often exhibit deficits in empathy and ToM and elevated levels of alexithymia (17–19), it remains unclear whether these traits extend to their unaffected male siblings. Brothers of women with AN are an understudied group that could provide valuable insights into whether these psychological traits represent broader familial markers of vulnerability or are specific to the disorder (7). Moreover, recent research conducted during the COVID-19 pandemic has revealed that unaffected sisters of women with AN exhibited intermediate emotional response to stress, suggesting that siblings may experience subclinical levels of emotional dysregulation (20, 21). This finding raises important questions about whether brothers—despite being less frequently diagnosed with AN—might also display subtle emotional-cognitive differences, such as alterations in empathy, ToM, and alexithymia. If so, this could indicate that familial risk factors for AN extend beyond gender, potentially affecting male siblings as well. Similarly, the literature has already reported the presence of endophenotypic features in AN among unaffected sisters, such as self-shifting abilities and central coherence (22–24). Taken together, these findings support the hypothesis that cognitive-emotional traits related to AN may be heritable and expressed across the family system, even in individuals not affected by the disorder itself. Psychological traits such as empathy, ToM, and alexithymia may not only reflect core dimensions of AN psychopathology but may also serve as cognitive-affective endophenotypes—inheritable traits that mediate the pathway between genetic risk and clinical expression (25, 26).

Descriptive statistics are presented in Table 1 , including paired comparisons between the bAN and AN groups, as well as between bAN and GP. Wilcoxon signed-rank tests showed that eating psychopathology differed significantly between bAN and their affected sisters. Additionally, specific differences in emotional features were observed across both comparisons. The bAN group exhibited lower scores on performance-based measures where lower scores indicate poorer outcomes (e.g., ToM and empathy tasks) and higher scores on self-report scales where elevated scores reflect greater emotional dysfunction (e.g., alexithymia subscales) when compared to the GP group. In contrast, the opposite pattern was generally observed when comparing bAN with their affected sisters.

Notably, the bAN group significantly differed from the GP group on several scales: EQ-SS (r = 0.361), TAS-DIF (r = 0.363), TAS-EOT (r = 0.360), TAS-20 Total (r = 0.450), RMET (r = 0.302), SET-IA (r = 0.291), SET-CI (r = 0.594), and SET-EA (r = 0.381). Comparisons between bAN and AN also revealed differences in SET-EA (r = 0.314), SET-EE (r = 0.313), SET-SS (r = 0.388), TAS-DDF (r = 0.348), TAS-DIF (r = 0.613), TAS-20 Total (r = 0.422), and eating psychopathology (all r > 0.700).

To assess group differences across cognitive and emotional variables, Kruskal–Wallis H tests were conducted on total and subscale scores across the three groups. The analysis revealed significant omnibus effects for several key measures. Pairwise comparisons using Bonferroni correction showed that both the AN and bAN groups scored significantly lower than controls on the SET-CI subscale. For SET-EA and SET-IA, the most pronounced differences were observed between the AN and GP groups. Additional comparisons across EQ and TAS-20 subscales—both significant and non-significant—are reported in Table 2 .

To assess multicollinearity among predictors in the logistic regression model, VIF values were calculated for all variables. All VIFs ranged from 1.08 to 2.69, well below the commonly accepted thresholds of 5 or 10, indicating no concerning collinearity. The highest VIF values were observed for the two alexithymia subscales—Difficulty Identifying Feelings (VIF = 2.64) and Difficulty Describing Feelings (VIF = 2.69)—which is expected given their conceptual and statistical overlap. However, these values remain within acceptable limits, supporting the inclusion of all eight predictors in the regression model without risk of distortion due to multicollinearity.

A binary logistic regression was conducted to assess whether specific empathy- and ToM-related variables significantly predicted group membership (bAN vs. GP). The model included eight predictors: RMET, SET-IA, SET-CI, SET-EA, EQ-EE, EQ-SS, TAS-DDF, and TAS-DIF. The overall model was statistically significant, χ²(8, N = 62) = 29.38, p <.001, explaining 50.3% of the variance (Nagelkerke R²). The only variable significantly associated with group membership was SET-CI, with lower SET-CI scores increasing the likelihood of belonging to the bAN group (B = 1.976, SE = 0.694, Wald = 8.112, p = .004, OR = 7.213, 95% CI [1.852, 28.097]). All other predictors, including RMET, EQ subscales, and TAS-20 subscales, did not reach significance. See Table 3 for details.

A sensitivity analysis was conducted by re-running the logistic regression model without the SET-CI predictor. The revised model remained statistically significant, χ²(7, N = 62) = 30.45, p <.001, explaining 35.4% of the variance and correctly classifying 80.6% of cases. In this model, TAS-DIF was the only significant predictor of group membership (B = 0.218, SE = 0.076, Wald = 8.12, p = .004, OR = 1.243, 95% CI [1.071, 1.444]). All other predictors remained non-significant.

This study explored social-cognitive and emotional processing in male siblings of individuals with anorexia nervosa, a population rarely investigated in the context of familial vulnerability to eating disorders. The findings indicate that bAN participants showed significantly lower performance than general population controls on several tasks assessing ToM and inferential reasoning. Additionally, they reported higher levels of alexithymia, particularly in difficulties identifying and describing feelings, reinforcing the view that emotional processing differences may not be limited to individuals with clinical eating disorders.

Among these variables, causal inference emerged as the most robust discriminator of group membership, with the strongest effect size and a significant predictive role in the logistic regression. While originally conceptualized as a control condition in SET paradigms (30), causal inference may engage domain-general reasoning and integrative functions—such as weighing contextual cues, mental state attribution, and sequencing of events—that are crucial for adaptive social cognition (32). These findings echo prior literature linking causal inference performance to broader neurocognitive vulnerabilities in conditions such as autism spectrum disorder and schizophrenia (33–35), and suggest that it may warrant greater attention as a potentially sensitive marker in transdiagnostic risk models.

The observation of group differences across other performance-based ToM tasks but not in all self-reported empathy questionnaires could underscore an important dissociation between subjective and objective social-cognitive measures. This may reflect known limitations of introspective access to empathic ability, particularly in populations with elevated alexithymia (36). Moreover, sex-related variability in affective reporting, with males often underreporting emotional difficulties or scoring lower on affective empathy (15), could attenuate group differences in questionnaire-based assessments. This distinction highlights the importance of combining multiple assessment modalities in the study of social cognition.

Notably, the presence of specific cognitive impairments—particularly in theory of mind and causal inference abilities—in siblings without a history of eating disorders echoes patterns commonly observed in individuals with AN (17, 18), and adds to the literature on potential familial or trait-like markers of vulnerability to the disorder. While previous studies have identified intermediate cognitive-affective profiles in unaffected sisters (22, 24–26), investigations of male siblings remain scarce. These results suggest that familial liability may manifest through subtle neurocognitive signatures—particularly those involving reasoning and emotion integration—even in males who do not develop overt psychopathology. However, the cross-sectional nature of the study precludes conclusions about causality or developmental sequence, and longitudinal designs will be necessary to determine whether these traits predict future clinical outcomes.

Another consideration relates to the possibility of compensatory mechanisms. The relatively preserved scores on some self-report measures could reflect successful cognitive adaptation in unaffected siblings, despite underlying differences in processing styles. This is consistent with resilience literature suggesting that individuals at risk for psychiatric disorders may develop alternative strategies to maintain social or emotional functioning (37). Alternatively, the findings may reflect latent vulnerabilities that require interaction with environmental or stress-related triggers to become clinically relevant.

In an exploratory sensitivity analysis excluding SET-CI from the regression model, TAS-DIF emerged as a statistically significant predictor of group membership. This result may suggest that, when causal inference is not accounted for, difficulties in emotional awareness contribute more prominently to the differentiation between bAN and control participants. Although SET-CI remained the only significant predictor in the full model, this secondary finding highlights the potential relevance of alexithymic traits in the broader context of familial vulnerability to AN. Given the limited sample size and the exploratory nature of this analysis, these results should be interpreted with caution. Nevertheless, the finding are consistent with previous research linking alexithymia to emotional dysregulation in AN (13) and aligns with the view that alexithymia may not only co-occur with AN but could also represent a transdiagnostic vulnerability marker (38, 39), even in the absence of overt pathology. Finally, these results contribute to the growing recognition that eating disorders in males remain underdiagnosed and under-researched (2). While the clinical presentation of AN may differ by sex, the presence of similar cognitive-affective vulnerabilities in male relatives reinforces the need to move beyond gendered assumptions in both research and prevention. Identifying subtle familial markers—such as impairments in causal inference or emotional processing—may support the development of more targeted early detection strategies, particularly in populations that do not typically present with classical symptom profiles.

This study contributes to an evolving understanding of the cognitive-affective architecture underlying familial vulnerability to AN. The presence of inferential reasoning impairments and elevated alexithymic traits in unaffected male siblings suggests that these individuals may express subclinical manifestations of traits associated with AN. These findings lend support to the conceptualization of cognitive-affective dimensions as transdiagnostic markers, extending beyond symptomatic expression and diagnostic boundaries.

Rather than focusing solely on global empathy or traditional markers of social cognition, these results emphasize the relevance of more nuanced reasoning processes—such as causal inference—in shaping vulnerability to psychopathology. Future research should adopt longitudinal and multi-informant designs to better understand the developmental trajectories of these traits, as well as their interaction with gender, environment, and neurocognitive functioning. By identifying specific cognitive-emotional profiles in at-risk individuals, we move closer to developing precision-based approaches to prevention and early intervention in eating disorders.