Qualitative Differences in Attribution of Mental States to Other People in Autism and Schizophrenia: What are the Tools for Differential Diagnosis?

Some symptoms in both disorders have common features, and the clinician is not always able to refer these symptoms easily to one or the other disorder. For example, social difficulties and pragmatic and grammatical language impairment are characteristics of both conditions (Chisholm et al., 2015; De Crescenzo et al., 2019; Solomon et al., 2011). Language deficits typical of autism appear similar to those found in prodromal symptoms of schizophrenia (De Crescenzo et al., 2019; Owen et al., 2011). In addition, formal thought disorder is similar to tangential thought, formal language, and the focus on restrictive interests shown by people with autism. Likewise, a lack of verbal initiation and poverty of speech content seems to be common in both disorders (De Crescenzo et al., 2019; Solomon et al., 2011).

Even the use of clinical tests for the diagnosis of ASD does not always help the clinician in the differential diagnostic process. Moreover, diagnostic tools designed for adulthood are limited (Maddox et al., 2017).

The standard diagnostic test for autism is the Autism Diagnostic Observation Schedule—Version 2 (ADOS-2; Lord et al., 2012), but despite its widespread use, it is worth noting that ADOS-2 is not specific enough when used in adolescents or adults with psychotic onset (Frith & Happé, 2005; Keller & Bari, 2019; Pino et al., 2018). Social withdrawal and social communication difficulties in ASD usually overlap with negative SSD symptoms or catatonic behaviour, thereby leading to diagnostic errors (Bastiaansen et al., 2011; Frith & Happé, 2005). Several studies have shown that ADOS-2 Module 4, usually intended for adolescents and adults with fluent language skills, might not correctly distinguish between ASD and psychosis, producing a high rate of false positive (Bastiaansen et al., 2011; de Bildt et al., 2016; Hus & Lord, 2014; Maddox et al., 2017; Mazza et al., 2010). Indeed, some items in the Communication (conversation, emphatic or emotional gestures) and Reciprocal Social Interaction domains of ADOS-2 Module 4 (unusual eye contact, facial expression directed towards others, empathy towards/comments on others’ emotions, quality of social overtures, quality of social responses, amount of social reciprocal communication) could reflect negative symptoms of psychosis rather than typical features of ASD.

A tool that could help the clinician during the diagnostic process is the Autism Diagnostic Interview—Revised (ADI-R; Lord et al., 1994), a semi-structured interview administered to caregivers in order to collects developmental information on the patient’s early life. However, ADI-R could lead to false negative in Asperger’s syndrome because symptoms are detected later, upon entering school; moreover, use with adult patients may be limited by the age of the parents and the quantity and quality of memories relating to the child’s early life and developmental milestones (Keller & Bari, 2019).

Based on recent literature and on diagnostic criteria of DSM-5 (American Psychiatric Association, 2013), the problem of a correct diagnosis, especially when the clinician faces an adolescent or adult with social withdrawal, anxiety, compulsions and/or stereotyped thinking, remains a very important clinical question.

The relationship between ASD and SSD is consistent with a neurobiological model that suggests a common basis for ASD and SSD, with several genetic alterations or neuroimaging investigation (Biamino et al., 2016; de Lacy & King, 2013; King & Lord, 2010). Neuroimaging studies showed common abnormalities in gray matter volumes and brain activations (Cheung et al., 2010; Pinkham et al., 2008; Sugranyes et al., 2011; Yoshihara et al., 2020). Areas with the greatest overlap are those correlated with Social Cognition (SC). SC is a complex set of cognitive abilities involved in the processing and interpretation of social stimuli (Bishop-Fitzpatrick et al., 2017; Mazza et al., 2010, 2017; Pino et al., 2017, 2020). Sugranyes et al. (2011) suggested that both ASD and SSD were characterised by media-temporal and ventrolateral prefrontal hypoactivity during SC tasks. Similarly, Pinkham et al. (2008) found reduced brain activity in the amygdala, fusiform gyrus and ventrolateral prefrontal cortex in individuals with autism and paranoid schizophrenia during trustworthiness of human faces tasks (Eack et al., 2017). Comparable reductions in superior temporal sulcus and ventromedial prefrontal cortex activity were found by Ciaramidaro et al. (2015) during a Theory of Mind (ToM) task. ToM is a main component of SC which can be summarised as the ability to understand mental and emotional states of other people (Mazza et al., 2014, 2017; Pino et al., 2020).

From a neuroimaging perspective, a voxel-based meta-analysis observed an overlap between clusters of brain alteration in ASD and SSD, involving areas related to the sense of the self, interoception, emotions and SC (Cauda et al., 2017). Interestingly, in ASD and SSD, brain alterations do not distribute randomly but rather follow network-like patterns of co-alterations, involving highly connected nodes.

Deficits in ToM abilities are considered mechanisms underlying social dysfunctions of both disorders. On the other hand, differences in ToM between ASD and SSD are delineated during specific phases of development; indeed, Frith (1995, 2004) suggests that, in contrast to ASD individuals, ToM abilities in people with psychosis develop normally but become impaired following the first psychotic episode (Bliksted et al., 2017; Mazza et al., 2001). People with psychosis may lose an ability they once had (Mazza et al., 2001), and that makes them socially similar to people with ASD.

However, the literature provides opposing evidence about the hypothesis that ToM is similarly compromised in both disorders (Chung et al., 2014; Deste et al., 2020; Fernandes et al., 2018; Morrison et al., 2017; Pinkham et al., 2019; Sasson et al., 2011; Tin et al.,, 2018) and that it could be disorder-specific (Brighenti et al., 2018; Colle et al., 2019). According to Morrison et al. (2017), the two disorders may differ in the nature and mechanisms of ToM deficit, but these differences could remain unidentified due to a lack of direct comparisons between ASD and SSD via systematically matched designs.

Some studies (Bowler, 1992; Couture et al., 2010; Craig et al., 2004; Murphy, 2006; Pilowsky et al., 2000) found that ASD and SSD patients did not differ in ToM verbal and visual tasks.

Tin et al. (2018) showed that ASD individuals, when directly compared with SSD patients, performed poorer in verbal but not in visual ToM. The authors conclude that ASD individuals shared similar but more severe ToM impairments than SSD patients did.

Martinez et al. (2019) showed that individuals with SSD were significantly impaired in the attribution of intentions to others (ToM), and that they did not differ from persons with ASD in the animated shapes task.

According to Frith (1992), social difficulties in ASD are correlated with poor mentalising ability; in particular, individuals with ASD are unable to represent the mental states of other people (Frith, 1992; Pickup & Frith, 2001; Tin et al., 2018). On the contrary, hypermentalising and an erroneous representation of others’ mental states are typical of SSD (Frith, 1992; Pickup & Frith, 2001; Tin et al., 2018). In line with what Frith claimed in 1992, many studies have suggested that social impairments in the two disorders may manifest in different ways: ASD are characterised by ‘under-mentalising’ and SSD tends towards ‘over-mentalising’ in ToM tests. Indeed, SSD patients have a higher degree of intentionality in their interpretations (Bliksted et al., 2016; Crespi & Badcock, 2008; Crespi et al., 2010; Martinez et al., 2019; Morrison et al., 2017). Fernandes et al. (2018) highlight that the unclear results reported in the literature about differences between ASD and SSD in terms of ToM performance could be related to a lack of tools able to discriminate between the disorders. The difficulty of researchers and clinicians in understanding the different performance of the two groups in ToM tests may be due to the lack of sensitivity of the tests used (Baksh et al., 2020; Brewer et al., 2017; Kuo et al., 2019; Morrison et al., 2019; Pino et al., 2020). Moreover, the development of ToM test for ASD individuals reports limitations, such as the fact that ToM tasks are performed in experimental settings that are different from the social situations of real life.

Hutchins et al. (2012, 2016) highlighted that ToM tasks require explicit and applied skills. The authors define explicit ToM competence as conceptual and logical knowledge that is limited to experimental contexts. Instead, applied ToM competence would be characterised by the ability to apply ToM knowledge to address social situations presented by the real world (Hutchins et al., 2016). Explicit and applied ToM competencies may be dissociated; for example, although adults with Asperger’s syndrome perform explicit tasks, they may use logical rather than mentalistic processes to provide the correct response (Bowler, 1992; Hutchins et al., 2016).

In addition, several ToM tests are based on language abilities. According to the literature, ToM skills and development of language are related (Apperly et al., 2009; Astington & Jenkins, 1999; Milligan et al., 2007; Pino et al., 2018; Sivaratnam et al., 2012). Several studies highlighted that individuals with ASD and SSD exhibit atypical cognitive, language and social functioning (Mazza et al., 2008; Solomon et al., 2011). Moreover, some studies showed that deficit in ToM abilities may be associated with sematic-pragmatic language, which is compromised both in ASD and SSD (Bambini et al., 2016; Deliens et al., 2018; Mazza et al., 2008; Pawelczyk et al., 2020).

There is enough evidence, since the pioneer research of Happé (1994), to hypothesise that individuals can use two ways to understanding the intention of others: a ‘mental’ reasoning and a ‘physical’ reasoning. The first involves mental states because it refers to feelings, thoughts, desires, dispositions and traits. The second is based on non-mental events such as the action of objects, physical events, physical appearances and outcomes.

Based on these assumptions, the aim of our study is: (a) to evaluate and describe differences in the style of reasoning used to solve Advanced ToM tasks (Blair & Cipolotti, 2000; Happé, 1994) between typically developing (TD) individuals and clinical populations (ASD and SSD); and (b) to understand whether the reasoning style used for the mentalising task differs between ASD and SSD. From a translational perspective, the overall goal is to offer clinicians an efficient way to identify items of the ToM test that discriminate between ASD and SSD.

One hundred and sixty-six subjects participated in this study, classified in three groups:

Details about socio-demographic and clinical data of all participants are reported in Table 1. No linguistic, cultural or racial differences were present. We included only individuals who scored above the 25th percentile on Raven Progressive Matrices (Raven, 1938) or who had an Intellectual Quotient (IQ) above 85. All the participants were native Italian speakers and gave informed consent to participate. The data were collected from January 2014 to December 2019.

Demographic data were obtained from each participant before the administration of A-ToM.

A-ToM was administered to each participant individually by a professional psychologist in a quiet room without any distractions; participants’ responses were registered using paper and pencil.

After the administration of the task, each justification question was categorised by six professional psychologists individually according to the three categories described above (mental state, physical state and incorrect response); afterwards, the overall concordance among raters was calculated (Fleiss K = 0.77; p < .001, showing high agreement among raters), and any disagreement was discussed until an agreement was reached.

The study was designed and conducted according to the principles established by the Declaration of Helsinki and informed consent was obtained from each participant before the test was administered.

A multinomial logistic regression model was performed, with A-ToM stories’ scores as independent variables and study group (TD, ASD and SSD) as the dependent variable, to estimate how likely stories are to predict the reasoning style. We used TD as the baseline in order to explain how A-ToM stories predict the reasoning style of ASD or SSD. We considered the three levels of response (0, 1, 2) as the modalities of an ordered discrete variable (progressing from the least to the most correct response, i.e. 0 to 1 to 2) so that the model prediction could reflect the correctness of the responses. Thus, regression beta coefficients indicate an increase or decrease in the risk of diagnosis based on the correctness of the response. A multinomial logistic regression model was performed as an omnibus test to understand which stories are suitable to significantly differentiate between typical and clinical populations (ASD or SSD) based on response style. In fact, if a story had significant results, it indicates that response style (incorrect, physical or mentalistic) is crucial for classifying the individual in one group or another; that is, how the participant responds to the story increases or decreases the relative odds of being in one of the clinical groups rather than in the typical group, showing a valid diagnostic performance, thus indicating that the style of response to that story could aid discrimination between groups. Between-group differences were further evaluated through contingency tables. In fact, we performed a chi-square analysis (or extended Fisher’s exact test as needed) of contingency tables for each of the 13 A-ToM stories, to verify the association between A-ToM scores and groups and to provide a better description of the differences. Moreover, a direct comparison between ASD and SSD was performed in order to describe differences between the two clinical group. Single stories are reported by story number followed by the story’s category; for example, A-ToM item 1, which represents the ‘Pretend’ category, is reported as 1_PRETEND. The analysis was performed using SPSS.

Both clinical groups had difficulty responding to stories that investigate the construct of Pretend, as compared to the TD group. Pretend is the ability to create imaginary events and to establish alternative identities for objects and people (Bruner, 1972; Rutherford & Rogers, 2003). According to Leslie (1987), pretend play is an early manifestation of the meta-representational abilities necessary for the development of a ToM. As suggested by Lillard (1993), pretend play is based on the ability to hold two mental representations in the mind (Kang et al., 2016; Leslie, 1987; Lillard, 1993). While the first representation reflects the state of the real world (perceived situation), the second reflects the pretend situation (Kang et al., 2016). Our results showed that participants in the SSD group were significantly more likely to respond incorrectly to both pretend stories (1_PRETEND and 13_PRETEND) than participants in the ASD group. Specifically, participants with SSD showed a tendency to attribute bad intentions to characters who are playing pretend or distorting reality (e.g. Emma is a liar; Emma teases Katie; She doesn’t want to give the banana to Katie; Philip is wrong). This result could be related to the incapacity of SSD individuals to use contextual information to make appropriate inferences about other people’s mental states (Frith, 1992; Langdon et al., 2001).

On the other hand, even when ASD participants answer the justification question incorrectly, they remain anchored to the specific context (e.g. Children don’t have much imagination; In fantasy everything is possible; Philip says it just for fun). In addition, our results highlighted a higher probability that the ASD group would respond using a mentalistic style in their responses than the SSD group (e.g. She’s pretending that banana is a phone; Philip pretends to be a pirate). According to the literature, most children with ASD exhibit an impairment of the ability to ‘play pretend’ (Gallagher & Varga, 2015; Happé, 1995; Happé & Frith, 2014; Jarrold, 2003; Kang et al., 2016; Leslie, 1991). However, it seems that many individuals with ASD may have an underlying ability to understand others’ actions of pretend, even if they fail to engage in spontaneous pretend (Jarrold, 2003, Jarrold et al., 1994; Kang et al., 2016; Kavanaugh & Harris, 1994).

Our clinical groups had significant difficulty, as compared to the TD group in responding to stories describing Persuasion. This construct requires understanding of a symbolic process that involves an attempt to bring about a change in attitudes, or to instil a particular belief in another person in an atmosphere of free choice (Lonigro et al., 2017; Perloff, 2010; Petty & Briñol, 2015). The central core of persuasion is the understanding that other people have different mental states that can be changed (Lonigro et al., 2017). This process requires adequate ToM skills. Interestingly, this once again underlines that SSD individuals tend to attribute bad intentions to the protagonist of the story. For instance, in regard to 2_PERSUASION, individuals with SSD are more likely to believe that the woman intended to drown the kittens, thereby attributing negative intentions to her. This paranoid ideation reflects the tendency of some individuals with schizophrenia to over-attribute intention to others (Green et al., 2015; Scherzer et al., 2012); negative symptoms and some of the positive symptoms of SSD, like paranoia, can be explained as impairment in the initiation and monitoring of willed actions. This reflects a difficulty in knowing whether or not an action was one’s own.

Our attention was focused on 11_DOUBLE_BLUFF: it is the only story that evaluates third-order ToM (Happé, 1994). In third-order ToM, the meta-representational level of recursivity improves until the third embedded belief (e.g. ‘I think that you think that he/she thinks that another person thinks…’ Happé, 1994; Valle et al., 2015). According to Valle et al. (2015), the third-order false belief task requires you to predict the character’s behaviour after attributing to him a mental state. The complexity of the recursive thinking of the third-order ToM can directly detect the implicit meta-representative mechanisms necessary to attribute a mental state and predict behaviour based on attribution. It is believed that ToM abilities continue to develop during adolescence and, in particular, during early adulthood, when higher-order recursive thinking emerges (Valle et al., 2015). Several studies have investigated how the development of ToM is linked to cognitive skills such as language, memory and executive functions; this relationship has been found in both clinical populations (SSD and ASD) and the TD population (Apperly et al., 2009; Demetriou et al., 2018; German & Hehman, 2006; Leung et al., 2016; Marinopoulou et al., 2016; Mutter et al., 2006; Valle et al., 2015; Wang et al., 2018).

Nonetheless, it seems that accuracy in the use of recursive thinking is closely related to social experience and the ability to predict the behaviour of others (Valle et al., 2015). As highlighted above, psychotic disorders mostly emerge during adolescence and early adulthood, and this coincides, therefore, with the development of higher-order ToM skills. According to Wang et al. (2018), it is not surprising that there is an impairment of ToM in psychotic patients (Mazza et al., 2010). In the 11_DOUBLE_BLUFF, one-third of SSD participants miss the comprehension question because the social scenario is too complex (see Table 2).

The SSD group shows a tendency to miss the justification question in all the constructs. Instead, the ASD group shows better performance by using mostly mental reasoning. This trend is not confirmed for the story that investigates the joke construct, where the ASD group shows a tendency to use physical reasoning. Verbal irony is a form of figurative or non-literal language in which the intended meaning of the speaker is frequently different from the literal meaning of his or her words (Pexman et al., 2011). Understanding ironic jokes (Sullivan et al., 1995) involves complex inferences about the speaker’s mind, requires attention to several social cues and is related to the ability to attribute at least second-order mental states (Deliens et al., 2018; Pexman, 2008). In stories involving a joke, the speaker knew that the listener knew the truth (a true second-order belief) and did not expect the listener to believe what was said. The literature has shown that individuals with ASD have deficits in the interpretative skills required for figurative language comprehension, compared to their TD peers (German & Hehman, 2006; Happé, 1994; Losh & Capps, 2006). Moreover, when presented with an ironic scenario, the ASD person tends to use explanations that involve reinterpreting the context to make the literal meaning fit (MacKay & Shaw, 2004; Pexman et al., 2011).

Irony is a type of figurative language that is used to convey messages in an indirect and sometimes amusing way (Saban-Bezalel et al., 2019). Irony is widely used in everyday life (Gibbs, 2000), and it helps to strengthen social bonds (Clark & Gerrig, 1984; Saban-Bezalel et al., 2019). Understanding irony requires the ability to go beyond literal language through complex mental representations (Saban-Bezalel et al., 2019). We found that our clinical groups had difficulty responding to stories that investigate irony. In the 3_IRONIC_JOKE, where the protagonist calls a dog an elephant, about four out of five ASD participants use a physical justification, for example, ‘the dog is big like an elephant’ or ‘because Giovanni sees a dog as big as an elephant’, evidencing little understanding of the mental states involved (ironic joke). This is in line with the literature, which highlights how ASD individuals perform better in metaphorical tasks, where the ability to accomplish first-order ToM tasks (inferring a person’s mental state, e.g. what he/she thinks) is required and a literal interpretation is sufficient (Happé, 1994; Sperber & Wilson, 1986). As for the SSD group, they showed high percentages of incorrect responses. In both ironic joke stories at least one out of three did not provide a correct mental or physical response to the justification question; examples of incorrect responses included ‘Because Giovanni was an elephant’, ‘Because he had this feeling’, for 3_IRONIC_JOKE, and ‘Hair is cut short to make other people laugh’, ‘to keep the grass from being mown’ for 10_IRONIC_JOKE (see Online Appendix). The qualitative analysis of their answers shows that they tend to use language characterised by reduced semantic coherence and syntactic complexity as well as by reduced vocabulary (Andreasen & Grove, 1986; Corcoran et al., 2018). Previous studies concerning SSD have reported that ToM difficulties are associated with dysfunctions in the processing of visual-perceptual and linguistic context (Dwyer et al., 2019; Schenkel et al., 2005), but they do not clarify the specific modes of attribution of mental states that the subjects use to understand and predict the behaviour of others.