Contrasting social knowledge and theory of mind patterns in adults with personality disorders, schizophrenia spectrum disorders, and healthy controls

ABSTRACT Introduction Personality disorders (PD) and schizophrenia spectrum disorders (SSD) are distinct conditions displaying common symptoms, like impairments in social cognition, that make them hard to distinguish, especially in severe cases. To date, few studies have compared theory of mind skills in these two disorders, and none have compared social knowledge skills. This study aims to compare the social cognitive abilities of patients with these conditions. Method Non-parametric analyses of covariance were used to compare severe PD patients (n = 37), SSD patients (n = 44), and healthy controls (HC; n = 49) on the Social Knowledge Test and two measures of theory of mind: the Reading the Mind in the Eyes Test and the Combined Stories Test, which incorporates items from various widely used tests. Results While no significant group differences were found on the Social Knowledge Test, SSD patients performed lower than the HC group on both theory of mind tests. PD patients only had lower performance than the HC group on specific items from the Combined Stories Test. Conclusions PD and SSD patients demonstrated distinctive patterns of social cognitive impairments, with items of greater complexity or with an affective orientation being the most discriminant for PD.


Introduction
Personality disorders (PD) and schizophrenia spectrum disorders (SSD) are two severe psychopathologies with distinct etiologies and treatment needs (Green et al., 2015;Skodol, 2018).Nevertheless, symptoms overlap in these two disorders; they notably share hallucinations, dissociative episodes, and paranoid ideations (Barnow et al., 2010;West, 2021).Therefore, misdiagnosis is frequent, especially in severe cases, which can lead to inappropriate treatment orientation and iatrogenic deterioration (Beatson et al., 2019).Altered social and interpersonal functioning is also characteristic of individuals with both PD and SSD (e.g., Wilson et al., 2017), and deficits in social cognition are now recognised as an underlying factor contributing to these difficulties (Herpertz & Bertsch, 2014;Pinkham, 2014).In this context, a deeper understanding of how social cognition impairments are observed specifically in each of these disorders is crucial.
Social cognition is a broad construct referring to all "mental operations that underlie social interactions, including perceiving, interpreting, and generating responses to the intentions, dispositions, and behaviours of others" (Green et al., 2008(Green et al., , p. 1211)).It comprises four core domains: emotional processing, attributional style, social perception and knowledge (SK), and theory of mind (ToM; Pinkham et al., 2014).The present study focuses on SK and ToM.
SK refers to the theoretical awareness of the social rules, goals, or roles expected in different social situations (Green et al., 2008).It is assessed by tasks in which the participants are given contextual information without reference to specific characters or reactions (Lavoie et al., 2013).SK is essential to the adequate functioning of more complex social cognitive processes but has been less extensively studied than the other domains of social cognition (Achim et al., 2012;Green et al., 2008;Pinkham et al., 2018).Thus, no studies have investigated SK in patients with PD, while the results of a meta-analysis showed moderate SK deficits in SSD (Savla et al., 2013).
ToM represents the ability to infer and reason about the mental states of others, such as their beliefs, intentions, or emotions (Green et al., 2008;Pinkham et al., 2014).ToM is assessed by tasks in which participants must attribute a mental state to specific characters in given situations (contextualised ToM).Exceptionally, some tasks present stimuli complex enough, such as eye gaze expressing complex emotions, to assess ToM without additional contextual information (decontextualised ToM; Lavoie et al., 2013).ToM studies with PD samples have obtained equivocal results, with several studies showing ToM deficits, though others found preserved or even enhanced ToM capacities relative to healthy controls (HC; Bora, 2020;Chang et al., 2021;Németh et al., 2018).For instance, Arntz et al. (2009) showed superior performance of cluster-C PD compared to HC on a task assessing contextualised ToM.In their meta-analysis, Németh et al. (2018) found small ToM deficits in Borderline PD (BPD).More specifically, BPD participants were impaired on their contextualised ToM skills but showed no alterations for decontextualised ToM.Nevertheless, other studies have highlighted deficits in decontextualised ToM skills compared to HC (Richman & Unoka, 2015).
In contrast, several meta-analyses have demonstrated substantial deficits, with large effect sizes (d = 1.21 and d = 1.26), in both contextualised and decontextualised ToM in individuals with chronic schizophrenia or first-episode psychosis compared to HC (e.g., Bora et al., 2009;Sprong et al., 2007).These deficits were observed across various tasks and were distinctly associated with schizophrenia symptoms (Cayouette et al., 2023;Thibaudeau et al., 2023).ToM deficits characterised by exaggerated mental state attributions (overmentalizing) have been associated with positive symptoms of schizophrenia, while insufficient or impoverished mental state attributions (undermentalizing) have been linked to negative symptoms (e.g., Montag et al., 2011;Peyroux et al., 2019).
Studies comparing PD and SSD showed that SSD patients underperform PD patients on their decontextualised ToM skills and on tasks assessing higher-order contextualised ToM, i.e., ToM tasks that require making inferences about the mental states of more than one person (Murphy, 1998(Murphy, , 2006)).In contrast, while SSD patients showed more important ToM deficits than PD and HC, Vaskinn et al. (2015) found no significant differences in ToM abilities between BPD patients and HC.However, BPD patients tended to have lower scores.Nevertheless, Andreou et al. (2015) and Vaskinn et al. (2015) found that patients with BPD made more overmentalizing errors than HC, which contrasts with the more extensive pattern of deficits characterised by both overmentalizing and undermentalizing deficits found in SSD.Furthermore, Bertone et al. (2017) showed that SSD patients were impaired on tasks assessing contextualised and decontextualised components of ToM, while incarcerated men with Antisocial PD (ASPD) showed poorer performance for decontextualised ToM only.
In summary, the evidence suggests that PD and SSD present different patterns of social cognitive alterations, with SSD patients showing global deficits and PD patients experiencing specific impairments in more demanding social situations.Very few studies, however, directly compared ToM abilities between these populations.Those that have done so focused mainly on women with BPD or detained men, challenging the generalisation of the results, especially since alterations in ToM skills seem to vary depending on the particular PD.Finally, data on SK, an essential social cognitive domain linked to community functioning (Fett et al., 2011), has yet to be considered.Gaining a deeper understanding of the nature and specificities of impairments in different aspects of social cognition in both PD and SSD could facilitate differential diagnosis, improve the accuracy of screening procedures, and help to refer patients, especially those with severe impairments, to appropriate treatment, This study aims to assess and contrast patterns of SK and ToM deficits in severely affected PD and SSD patients compared to HC.From previous studies, we expected PD participants to show deficits in social cognition compared to HC, but to a lesser extent than SSD patients.Specifically, we hypothesised (a) significant group differences on tasks assessing more basic social cognitive processes such as SK, on which SSD patients should show poorer performance than PD and HC groups; (b) significantly poorer performance of SSD patients and slightly poorer performance of PD patients compared to HC on the task assessing decontextualised ToM ability; and (c) significantly poorer performance of PD and SSD patients compared to HC on the task assessing contextualised ToM, with SSD patients showing lower scores compared to PD.The study also explores which variables are more useful in distinguishing SSD and PD patients.

Sample
The study includes 130 participants (M age = 37.02; SD age = 13.11;53.8% women) from three different samples.See Table 1 for descriptive statistics for each sample.Participants in the PD group (n = 37) were recruited from a clinic specialised in the evaluation and treatment of severe PD.Common PD diagnoses were BPD (n = 12), narcissistic PD (n = 5), Schizoid PD (n = 1), PD not otherwise specified (n = 2), and mixed PD, i.e., patients who met criteria for more than one PD, usually cluster B disorders in the present sample (n = 17).Only one participant in the PD group was not taking any medication; 81.10% were taking antidepressants, and 78.38% were taking more than one medication, the most common being antipsychotics and anxiolytics.
Participants in the SSD group (n = 44) were all outpatients recruited through specialised teams within the public health system and community organisations.Diagnoses included schizophrenia (n = 21), schizophreniform disorder (n = 2), schizoaffective disorder (n = 12), delusional disorder (n = 2), and psychosis not otherwise specified (n = 8), with 25% of participants who had a recent onset of psychosis (< 5 years).According to the remission criteria established by Andreasen et al. (2005) for the Positive and Negative Syndromes Scale (PANSS; Kay et al., 1987), 43.18% of the participants in the SSD group were remitted at the time of the study, although information was unavailable regarding the six months criterion (see complete PANSS statistics in Supplementary Table 1).
Participants in the HC group (n = 49) were recruited through ads in local stores, social media, and by word of mouth.They were excluded if they were undergoing psychoactive pharmacological treatment, had a psychiatric illness, or had a first-degree relative with a psychotic disorder.
Additional exclusion criteria were applied for all groups: the presence of a neurological disorder, an intellectual disability, 1 or if they did not speak French as their first language. 2Participants in the SSD group were excluded if they had a history of traumatic brain injury or if they presented with drug or alcohol dependence.For the PD sample, a history of traumatic brain injury with no lasting effects or a mild substance-related addiction was tolerated, considering the high prevalence of these issues in the PD population (Liao et al., 2012;Walter, 2015).The study was ethically approved by the Intersectoral ethics committee in neuroscience and mental health of the Integrated university health and social services center of the Capitale-Nationale, and all participants gave informed consent to participate in the research.

Social Knowledge Test (SKT)
The SKT, developed by Achim et al. (2012), assesses SK by asking participants to determine what emotions would be felt by most people in 14 hypothetical situations.Since no specific characters, actions, or verbalizations are presented, participants can rely only on the knowledge they previously possessed about each type of situation.This distinguishes the SKT from tasks assessing ToM, in which specific characters perform actions or make statements in addition to the presented context.According to a validated correcting grid (Achim et al., 2012), correct answers are awarded 1 point.Psychometric studies showed good validity and reliability, with moderate internal consistency (Achim et al., 2012;Thibaudeau et al., 2018).In the present study, for the combined sample, the Kuder-Richardson-20 coefficient (KR-20) was calculated (KR-20 = .40). 3   Reading the Mind in the Eyes Test (RMET) The RMET (Baron- Cohen et al., 2001;adapted by Pinkham et al., 2018) assesses decontextualised ToM.The test comprises 36 gray-scale pictures of the eye region presented on a computer.For each image, participants must choose from a list of three distractors and one correct answer, without a time limit, the word that best illustrates the mental state in the picture.The definition of each word is provided in this version of the Eyes test (Pinkham et al., 2018).Each correct answer is awarded a point.In the Social Cognition Psychometric Evaluation Study (SCOPE), this version of the test showed acceptable psychometric properties, with adequate internal consistency (the Cronbach alpha coefficient [α] = .75for SSD patients) and is sensitive enough to enable discrimination between SSD patients and HC (Pinkham et al., 2018).The present study calculated the Cronbach alpha for the combined sample (α = .40). 4   The Combined Stories Test (COST) The COST is a validated contextualised ToM task (Achim et al., 2012;Thibaudeau et al., 2018), initially developed by combining stories adapted from four ToM tests, namely the Hinting task (Corcoran et al., 1995), the False Belief Task (Baron- Cohen, 1989), the Faux pas task (Baron- Cohen et al., 1999), and the Strange Stories Test (Happé, 1994), as well as some stories developed by Achim et al. (2012), for a total of 30 stories.Participants are asked to read aloud the written stories and answer open questions targeting the character's mental state.The text is accessible to minimise the influence of memory load.The COST includes 26 questions assessing ToM (COST ToM; α = .74for the combined samples).It also contains control questions assessing general abilities to link a mental state to a behaviour, representing more basic ToM skills (COST first order inference; three questions), questions about details of the stories to control for memory or attentional deficits (COST memory/attention; 29 questions), as well as non-social reasoning (NSR) questions assessing general reasoning capacities that require drawing inferences about physical causalities that does not involve the actions or mental states of the characters (COST NSR; six questions; α = .61).See Table 2 for a more detailed description and classification of the questions.Only the NSR score was used in the present study from the control questions.The answers from the COST were scored on a 0-2 scale by raters working with a validated correction grid.The COST has previously shown excellent validity (Achim et al., 2012) and good test-retest reliability (Thibaudeau et al., 2018).

Statistical analyses
Analyses were conducted using the IBM Statistical Package for Social Sciences Statistics (version 28).Since significant group differences were observed in gender between the SSD group and the PD and HC groups (see Table 1), and because it is known that social cognition skills are higher in women (Zaroff & Ku, 2015), group differences for the SKT, RMET, and COST scores were assessed with gender as a covariate.Non-parametric univariate analyses of covariance (ANCOVA) were used since the variables were not all normally distributed and because there was heterogeneity of variance between groups.Cohen's ds were computed as a measure of effect size and interpreted according to Cohen's (1988) guidelines for effect magnitude (small: d = 0.2, medium: d = 0.5, large: d = 0.8).A power analysis conducted with the software G*Power 3.1 (Faul et al., 2007) indicated that a sample size of 130 participants detects an effect size of d = 0.70 with a power value of .95 and an alpha type I error rate of 0.05.

Group comparisons
As shown in Table 3, no statistically significant group differences were found on the SKT, with a small effect size observed between the SSD group and the PD and HC groups.Given these unexpected results for the SSD group, and since our sample presents high variability in illness duration, an additional correlation between the SKT total scores and illness duration was conducted for the SSD group and proved to be significant (r = −0.65,p < 0.001).For the RMET and the COST ToM, SSD patients had lower scores than the other two groups, with medium-large effect sizes, although only the difference with the HC group reached statistical significance.Finally, both clinical groups had lower scores than the HC group on the COST NSR questions, with medium-large effect sizes.

Supplementary analysis
As distinct patterns of ToM impairments were previously highlighted according notably to the specific tasks used (Bora et al., 2009;Németh et al., 2018), further analyses were conducted to examine if more precise patterns of group differences would emerge when looking at specific aspects of contextualised ToM (see Table 3).Thus, subscores  PD and SSD, PD and HC, and SSD and HC, respectively.b Difference between SSD and PD groups reach statistical significance for the RMET variable when considering non-social reasoning as a covariate, along with gender, F (2, 126) = 3.63, p = 0.03.c Difference between SSD and HC groups is no longer statistically significant for the false belief variable when considering non-social reasoning as a covariate, along with gender, F (2, 127) = 2.70, p = 0.07.* p < 0.05.**p < 0.01.***p < 0.001.
were computed for the COST items inspired by different original tasks.Non-parametric ANCOVAs were conducted with subscores as the dependent variables, group affiliation as the independent factor, and gender as a covariate.PD patients had significantly higher scores than the SSD group on the items inspired by the Strange Stories test and significantly lower scores than the HC group for the identification questions of the faux pas stories.There were no other statistically significant differences for the PD group.Although the SSD group had lower scores than the HC group for every variable tested, differences only reached statistical significance for the false belief questions, the strange stories questions, and the identification questions of the stories inspired by the Faux pas task.
Additionally, the previous significant statistical difference between HC and PD groups for NSR questions supports the hypothesis that cognitive abilities can influence performance on ToM tasks (Thibaudeau et al., 2020).Considering the presence of significant correlations between the NSR questions and social cognitive variables (r = .19,p = .033to r = .42,p < .001), the decision was taken to perform additional analyses adding NSR as a covariate, along with gender, to assess and control for the influence of non-social cognition on ToM abilities.When controlling for NSR and gender, only two variables showed results differing from the primary analyses.Firstly, the difference between the SSD group and the PD group became significant for RMET, while the difference between the SSD group and the HC group was no longer significant for the items inspired by the False Belief task.

Discussion
The present study was designed to compare the SK and ToM abilities of SSD and PD patients with HC participants as a control group.As expected, SSD and PD showed different patterns of social cognitive alterations.

Social knowledge
Although the results revealed the expected tendency that SSD patients had lower scores than both PD and HC participants, who performed similarly, no significant statistical differences emerged on the SKT.These results are consistent with the study of Achim et al. (2012), who found no significant differences between firstepisode psychosis patients and controls on the SKT.However, they are inconsistent with the results of the meta-analysis by Savla et al. (2013), who reported moderate SK deficits in SSD patients across seven previous studies.Nevertheless, the significant additional correlation obtained between SKT and illness duration for the SSD group indicated that patients in a less advanced stage of the disease have better SK skills.Additionally, over 40% of the participants in this study met Andreasen's et al. (2005) remission criteria, and 25% had a recent onset of psychosis (< 5 years).Altogether, these results suggest that SK abilities seem more preserved at the onset of a psychotic disorder and could deteriorate as the disease progresses or is in an acute phase.Thus, the high variability in illness duration for the SSD group and the high proportion of remitted patients in the present study may explain the lack of significant deficits.

Decontextualised ToM
As hypothesised, the SSD group exhibited global impairments in decontextualised ToM abilities (with a medium effect size) compared to the HC and PD groups.However, the difference with the PD group only reached statistical significance when controlling for NSR abilities, suggesting a lack of statistical power in the primary analyses or a small confounding effect of cognitive abilities on decontextualised ToM abilities for the SSD group.No statistically significant difference emerged between the PD and the HC groups for the RMET, which is consistent with a meta-analysis by Németh et al. (2018) and a review by Salgado et al. (2020).Indeed, identifying mental states from subtle perceptual cues, without additional contextual processing, may not require as much social reasoning abilities as other ToM tasks.Some authors even criticised the RMET itself, considering it an instrument assessing emotion recognition rather than ToM (Oakley et al., 2016).If so, the RMET may not be the most appropriate tool to highlight ToM impairments in PD.Nevertheless, some meta-analyses have found small to moderate deficits in emotion recognition in BPD (Hanegraaf et al., 2021) and ASPD (Marsden et al., 2019), and several have shown more extensive deficits in patients with schizophrenia (Bora et al., 2009).Thus, emotion recognition skills seem to be more impaired in SSD than in PD patients.Future studies could use more ecologically valid tasks or conditions to better highlight ToM impairments in PD (e.g., adding a time limit to resemble more closely to real-life interactions) and explore alternative measures to assess ToM abilities in PD.

Contextualised ToM
As expected, SSD patients underperformed the HC group on contextualised ToM.However, contrary to our predictions, their difference with the PD group did not reach statistical significance despite a medium effect size.Furthermore, while PD participants obtained lower COST scores than HC group participants, that difference was also not statistically significant.The lack of difference between the PD group and the other two groups may be explained by the variations in the specific ToM tests used (such as different types or levels of complexity of ToM processes), as frequently highlighted in the literature (Bora et al., 2009;Németh et al., 2018).To explore this avenue, supplementary analyses were performed to see if differences in contextualised ToM in the PD group would emerge when considering separately the COST items adapted from different initial tests.
The results from the aforementioned analyses indicated that SSD participants were affected on their ability to detect a faux pas (identification questions of the faux pas stories), to understand irony, white lies, or misunderstanding (strange stories questions), and to understand that a character holds a belief that differs from their own knowledge (false belief stories).However, the difference for the false belief stories was no longer statistically significant once controlling for NSR abilities.This suggests that cognitive abilities seem to play a greater role in the performance of the false belief stories for SSD.This finding aligns with the results of Fernandez-Gonzalo et al. (2014), who found that executive functions were related to second-order false beliefs.Additionally, Thibaudeau et al. (2020) proposed that characteristics of ToM tasks, such as the type of mental state, the type and the modality of answers, or the complexity of attributions, may influence the link between ToM and neurocognition.Nevertheless, further studies are needed to assess what specific characteristics of the false belief stories are influenced by NSR.Furthermore, while results indicated that participants from the SSD group are more impacted than the HC and the PD participants on their abilities to interpret irony, white lies, or misunderstanding (strange stories questions), the PD group showed similar, even better performance than the HC group on these tasks.These results align with those of Salgado et al. (2020), who reported no differences between BPD patients and controls in two studies using the Strange Stories test.Arntz et al. (2009) even found significantly enhanced capacity in patients with cluster C PD compared to healthy individuals.Current results and previous findings suggest that PD patients' ability to detect irony or white lies is not directly affected and that their interpersonal impairments are associated with other elements of ToM, like detecting a faux pas or a tendency to overmentalize.However, the Strange Stories test could be the more practical option when a differential diagnosis has to be made between SSD and severe PD.
Items inspired by the Faux pas task, precisely the identification questions (evaluating the ability to identify if a faux pas occurred and why it was a faux pas), were the only ones that allowed a significant distinction between PD and HC.These results coincide with those of Németh et al. (2018) and Salgado et al. (2020), who noted that it was particularly challenging for BPD patients to correctly identify when someone says something without realising that it may upset someone else and why (Baron- Cohen et al., 1999).The deficits in the faux pas stories suggest that impairments in PD are highlighted in specific contexts.Indeed, the faux pas stories describe complex social situations requiring making implicit inferences and integrating the perspective of multiple characters.However, the PD group only underperformed the HC group on the identification questions of the faux pas stories and presented no deficits on the false belief questions, meaning that when a faux pas was identified, PD participants could understand that it came from a false belief rather than from a malevolent intention.
An explanation may lie in the fact that the false belief questions (assessing whether the participant recognised that the two characters had different kinds of knowledge) evaluate the ability to infer cognitive mental states such as beliefs or knowledge (cognitive ToM), while the identification questions, which require appreciating the emotional impact of the faux pas, assess the ability to infer emotional mental states (affective ToM).Accordingly, a growing consensus supports a distinction between affective and cognitive ToM impairments (Harari et al., 2010;Németh et al., 2018).Indeed, ASPD is known to be particularly impacted in affective ToM abilities without showing any difficulties in understanding cognitive mental states (e.g., Chang et al., 2021;Dolan & Fullam, 2004).For their part, BPD patients seemed to be affected on both cognitive and affective ToM (e.g., Németh et al., 2018) or predominantly on the cognitive part of the faux pas stories (Harari et al., 2010).The high prevalence of mixed PD in our sample, with all PD diagnoses gathered in the same group, did not allow us to corroborate previous findings.Still, the identification questions, which assess affective ToM, seem particularly effective in highlighting some ToM impairments in PD.

Limitations and future directions
Some limitations of the present study need to be addressed.First, the sample was recruited from natural settings to maximise the generalizability of the results.Thus, it was impossible to match participants according to confounding variables known to impact social cognition skills, such as gender or neurocognitive abilities (Thibaudeau et al., 2020;Zaroff & Ku, 2015).The decision was also taken not to exclude any participants based on those variables to preserve the representativeness of the clinical populations.However, the gender effect was statistically controlled, as well as NSR in supplementary analyses.
Secondly, PD and SSD participants were undergoing pharmacological treatment at the time of the study, which is known to impact neurocognition and social cognition (Haime et al., 2021).However, the precise impacts of different medications on social cognitive abilities are not well understood and can vary from improvement to deterioration, depending notably on the dosage or type of medication used.Considering the significant variability of the pharmacological profiles of the participants in terms of dosage, type of medication, and the possible interactions between different medications (67.9% of the clinical samples taking more than one psychoactive medication), it became difficult to control for the medication effect on social cognition tests results.Therefore, the lack of control for medication calls for caution in the interpretation of the results.
Third, the sample size did not allow to perform separate analyses for specific PD diagnoses, even if different PDs may present distinctive patterns of social cognitive deficits.On the other hand, our sample presents many patients with mixed PD, which attests to its severity, but this is also common in populations with less severe PD (Widiger & Trull, 2007).This high comorbidity and high heterogeneity of symptoms for the same diagnosis has brought a paradigm shift towards more dimensional conceptualizations of PD, such as the Alternative DSM-5 Model for Personality Disorders (American Psychiatric Association, 2013).Thus, future studies might obtain more steady results focusing on personality functioning and personality-specific traits rather than diagnostic categories.
Finally, the modest psychometric properties of the RMET can lead to questioning its validity in assessing decontextualised ToM and its capacity to detect differences between PD participants and controls.Further studies involving PD would benefit from including more ecologically valid and sensitive tasks.

Conclusion
In summary, this study aimed to compare SSD and PD patients with HC on tasks assessing SK and contextualised and decontextualised ToM.The results revealed distinctive patterns of social cognitive alterations in patients with PD and SSD.To our knowledge, this is the first study to investigate SK abilities in both clinical groups, which showed equivalent SK abilities to controls.This was expected for the PD group but surprising for SSD patients.This contradictory finding suggests that SK deficits in SSD patients may emerge specifically in more symptomatic patients or patients with a longer duration of illness.ToM impairments in SSD are broad and observed across various types of stories.Furthermore, most of the difficulties observed for the SSD group are maintained

Table 1 .
Demographic characteristics for patients with schizophrenia spectrum disorders, personality disorders, and healthy controls.Personality disorders; SSD = Schizophrenia spectrum disorders; HC = healthy controls.a Post-hoc p values were adjusted with Bonferroni correction.

Table 2 .
Description of COST stories and questions.
Note: COST = The Combined Story Test; ToM = Theory of mind; NSR = Non-social Reasoning Questions.

Table 3 .
Group a Effect sizes for comparisons between