Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Sep 28.
Published in final edited form as: J Autism Dev Disord. 2022 Jun 25;53(9):3636–3647. doi: 10.1007/s10803-022-05620-0

Distinct Symptom Network Structure and Shared Central Social Communication Symptomatology in Autism and Schizophrenia: A Bayesian Network Analysis

Gloria T Han 1, Dominic A Trevisan 1, Jennifer Foss-Feig 2, Vinod Srihari 1, James C McPartland 1
PMCID: PMC10202012  NIHMSID: NIHMS1894569  PMID: 35752729

Abstract

Autism (ASD) and schizophrenia spectrum disorders (SCZ) are neurodevelopmental conditions with overlapping and interrelated symptoms. A network analysis approach that represents clinical conditions as a set of “nodes” (symptoms) connected by “edges” (relations among symptoms) was used to compare symptom organization in the two conditions. Gaussian graphical models were estimated using Bayesian methods to model separate symptom networks for adults with confirmed ASD or SCZ diagnoses. Though overall symptom organization differed by diagnostic group, both symptom networks demonstrated high centrality of social communication difficulties. Autism-relevant restricted and repetitive behaviors and schizophrenia-related cognitive-perceptual symptoms were uniquely central to the ASD and SCZ networks, respectively. Results offer recommendations to improve differential diagnosis and highlight potential treatment targets in ASD and SCZ.

Keywords: Autism, Schizophrenia, Social communication, Network analysis

Introduction

Autism spectrum disorders (ASD) and schizophrenia spectrum disorders (SCZ) are neurodevelopmental conditions with significant co-occurrence and overlapping symptom presentation. ASD is characterized by the core symptom domains of restricted interests and repetitive behavior (RRBs) and differences in social communication and interaction and SCZ is characterized by the core symptom domains of positive symptoms (e.g., delusions, hallucinations, thought disorder), negative symptoms (e.g., reduced affect, loss of pleasure, reduced emotional expression, and loss of motivation), and cognitive symptoms (e.g., difficulties with attention, working memory, and processing speed). Population-based and meta-analytic studies have indicated significantly higher incidence of SCZ in individuals with ASD (Chien et al., 2021), and higher levels of ASD symptomatology for individuals diagnosed with SCZ (De Crescenzo et al., 2019) compared to the general population. Recent work also suggests that gold standard diagnostic instruments for ASD reliably discriminate ASD from neurotypical adults but yield notable false positives for individuals with SCZ who do not meet criteria for ASD (Trevisan et al., 2020).

One possibility for this lack of diagnostic specificity is the presence of overlapping clinical features of ASD and SCZ. For example, differences in social communication and interaction represent a core symptom domain of ASD, and social withdrawal and flat affect are prominent negative symptoms in SCZ. Of note, a diagnosis of SCZ requires that the individual exhibits two or more core symptoms, one of which must include hallucinations, delusions, or disorganized speech. Though this additional qualification may facilitate differential diagnosis, topographical overlap between these features of SCZ and ASD symptoms are also observed in clinical contexts (Chandrasekhar et al., 2020). For example, ASD-specific RRBs in the form of repetitive or perseverative thinking may sometimes resemble paranoid or delusional thinking observed in SCZ. Such instances of topographic symptom overlap complicate differential diagnosis, which may hinder treatment conceptualization, access to appropriate treatment, and subsequent prognosis.

Evidence of comorbidity, symptom overlap, and heterogeneity within ASD and SCZ has motivated efforts to shift from categorical definitions of the two conditions to conceptualizing ASD and SCZ in terms of dimensional and transdiagnostic symptoms, consistent with the Research Domain Criteria (RDoC) framework (Insel et al., 2010). A shift from categorical to dimensional definitions entails the recognition that conditions emerge from a network of complex and causal interactions among symptoms (Borsboom & Cramer, 2013). This contrasts with traditional symptom checklist approaches to diagnosis wherein clinical conditions are defined as latent constructs that cause the emergence of observable symptoms. Though such approaches have persisted because of clinical utility, the mathematical derivation of latent constructs (i.e., clinical condition or diagnostic category) implies the assumption of local independence, which states that manifest variables (i.e., observable symptoms) are independent of each other within a latent construct (Sobel, 1997). This assumption is commonly violated, as observable symptoms within a diagnostic category are often interrelated (McNally, 2016). In the case of ASD, for example, engagement with one’s circumscribed interests may negatively impact attention to social cues and development of social competence. In this way, symptoms may be interrelated, and distinct clinical phenotypes (or diagnostic categories) may best be viewed as emerging from specific patterns of interactions among symptoms over development rather than from distinct sets of symptoms, per se.

In the network analysis framework (Borsboom & Cramer, 2013; McNally, 2021), the etiological assumption in which observable symptoms are caused by a common latent factor is relaxed, and symptoms are no longer considered measurements of a latent condition. Rather, the organization of and interaction between symptoms is what constitutes the condition. In practice, network analysis explores symptom interrelations (i.e., the conditional dependence structure of symptoms) and visually represents conditions as a network of symptoms, each represented by a node, connected by edges that reflect the strength of association between symptoms. In addition to identifying the overall network structure of symptoms, network analysis yields estimates of node “centrality” or the degree to which a given node is related and connected to other nodes within a network. Further, networks with symptoms representing more than one condition allow for the identification of “bridge symptoms” that connect symptoms from one psychiatric condition to another. Examination of network structures, central nodes, and bridge symptoms may then inform differential diagnosis, therapeutic targets, and pathways to comorbidity. For example, targeting symptoms with high centrality may alter the interplay between symptoms and mitigate the emergence or phenotypic presentation of the clinical condition. Characterizing differences in symptom networks may also offer insights for distinguishing disorders with overlapping symptomatology and targeting bridge symptoms may help to explain or prevent comorbidity.

Few studies have used network analysis to examine ASD and SCZ symptom overlap, though preliminary applications demonstrate promise for this approach. One study of 2469 college students utilized network analysis to examine the relationship between ASD and SCZ traits and found that negative schizotypal traits were strongly and positively correlated with ASD-related social and communication features, whereas positive schizotypal traits were negatively correlated with ASD traits (Zhou et al., 2019). This finding from a non-clinical sample is consistent with a framework for differentiating the two conditions in which ASD and SCZ demonstrate convergence in symptom presentation based on presence of “negative symptoms,” defined as absence of typical behaviors, and divergence based on the presence of “positive symptoms,” defined as the addition of a typical behaviors compared to individuals without a clinical diagnosis (Trevisan et al., 2020). While the nomenclature of positive and negative symptoms has been traditionally used to categorize SCZ symptoms, this framework demonstrates clinical utility in ASD and may inform procedures for more accurate and precise differential diagnosis (Foss-Feig et al., 2016).

Current Study

The current study aimed to inform differential diagnosis of ASD and SCZ by comparing the network structures of autism and schizophrenia symptoms in individuals diagnosed with ASD to individuals diagnosed with SCZ. To our knowledge, this is the first study to apply network analytic approaches to understand ASD-SCZ symptom overlap in a well-characterized clinical sample of individuals with confirmed diagnoses. Specifically, we examined between-group differences in symptom organization by (1) estimating separate symptom networks to examine the overall organization of autism and schizophrenia symptoms for individuals with confirmed diagnoses of ASD and SCZ, (2) testing whether the overall network structure (i.e., organization of symptoms) differs by diagnostic group, and (3) assessing differences in node centrality, or the degree to which a node/ symptom is highly connected to other nodes/symptoms in the symptom networks for individuals with ASD and SCZ. Informed by previous work (Kästner et al., 2015; Trevisan et al., 2020), we hypothesized that social communication symptoms would be central and highly connected to the other symptoms in the symptom networks of both diagnostic groups, while the centrality of SCZ-related positive symptoms (e.g., delusions, hallucinations) and ASD-related restricted interests and repetitive behaviors would be distinct to the SCZ and ASD groups, respectively. Differences in centrality findings between the two networks would help to inform differential diagnosis and potential treatment targets relevant to the two conditions. We also examined bridge symptoms that connect ASD symptoms to SCZ symptoms in the two symptom networks to inform potential pathways to comorbidity.

Methods

Participants

The study sample consisted of 92 subjects with ASD (n = 53) or SCZ (n = 39) aged 18–48. Participants were recruited for this study after seeking treatment, services, and/or research participation at the Yale Developmental Disabilities Clinic, the Specialized Treatment for Early Psychosis (STEP) Clinic, or the Yale Psychiatry department in New Haven, Connecticut. Diagnoses were confirmed by doctoral-level clinicians with extensive experience in both clinical populations. All participants were administered gold-standard diagnostic assessments of ASD and SCZ characteristics including the Autism Diagnostic Observation Schedule (ADOS-2) (Lord et al., 2012) and the Positive and Negative Syndrome Scale (PANSS) (Kay et al., 1987) and met DSM-5 criteria for ASD or SCZ. Clinician decisions were informed by multiple sources of information, including clinician interactions with participants during administration of diagnostic assessments and prior psychiatric histories obtained as part of the recruitment procedures.

Exclusion criteria included meeting DSM-5 criteria for both ASD and SCZ (n = 2), as the purpose of the current study was to assess differences in symptom network structure for individuals with confirmed, non-comorbid diagnostic status. Participants were also excluded from analysis if they had a full-scale intelligence quotient (IQ) score of less than 70 on the Wechsler Abbreviated Scale of Intelligence 2nd edition (WASI-II) to facilitate validity of the self-report measures. This study was approved by the Yale University School of Medicine Human Subject Investigation Committee.

Measures

Social Responsiveness Scale—Second Edition (SRS-2)

The SRS-2 (Constantino & Gruber, 2012) adult self-report form is a 65-item questionnaire assessing symptoms associated with ASD, with higher scores indicating higher levels of symptoms associated with ASD. The SRS-2 includes five subscales: social awareness (ability to recognize social cues), social cognition (ability to interpret social behavior), social communication (reciprocal communication in social situations), social motivation (motivation to participate in social interactions), and restrictive interests and repetitive behavior (circumscribed interests and stereotypy). For each item, respondents indicate agreement with each item on a 4-point Likert scale, rating their behavior over the past 6 months. T-scores for each subscale were used to represent the ASD-relevant symptom nodes in the network analysis.

Schizotypal Personality Questionnaire (SPQ)

The SPQ (Raine, 1991) is a 74-item self-report questionnaire designed to measure assessing symptoms related to schizotypy. Though it was originally designed and tested for its ability to discriminate schizotypal personality disorder, the SPQ has evolved as a dimensional measure of schizotypal traits relevant to individuals with SCZ. These traits reflect biological, cognitive, and social dimensions of schizotypal traits. The SPQ yields nine subscales, including ideas of reference, suspiciousness, magical thinking, unusual perceptions, no close friends, constricted affect, social anxiety, eccentric behavior, and odd speech. Each “yes” response is counted as one point and 9 subscale scores are computed as the total score for all items associated with each subscale. Subscale scores were used to represent SCZ-relevant symptom nodes in the network analysis.

Data Analytic Plan

All analyses were performed using R version 4.0.3 (R Development Core Team, 2020). The networks were constructed and visualized using the BGGM (version 2.0.3) package (Williams & Mulder, 2019, 2020) in R. Analyses were organized in the following manner: (1) estimation of diagnosis-specific networks, (2) estimation of node centrality and bridge strength in each network, and (3) comparison of overall network structure and individual edge weights.

Diagnosis-Specific Network Estimation

Clinical symptom networks were modeled using a Gaussian graphical model (Lauritzen & Wermuth, 1989) with Bayesian methods (Williams & Mulder, 2019, 2020). Best practices for network estimation and subsequent interpretation of network estimates require that the network demonstrates adequate stability (Epskamp et al., 2018). Stability is important to facilitate replication of network findings and comparison of symptom networks but is hindered by small sample sizes. Considering our modest sample size in the context of network analysis and that comparing the ASD and SCZ networks was the primary aim of the current study, we turned to a Bayesian approach for implementing network analysis that quantifies the degree of uncertainty of parameter estimates to account for instability due to sample size constraints. In brief, the Bayesian approach utilizes a prior probability distribution or “prior” that describes any knowledge or belief about some quantity or parameter. Then, by multiplying the prior with a “likelihood function” (i.e., the conditional probability of the observed data given the prior), posterior distributions of the parameters of interest are generated. As a distribution instead of point estimate, the posterior distribution is used to represent the parameter estimate and its uncertainty (i.e., width of the distribution). Described in more detail elsewhere (Williams & Mulder, 2020; Williams et al., 2020), applying the Bayesian framework to network estimation helps to mitigate concerns of sample size constraints and offers a more suitable method for estimating and comparing the overall network structure and specific aspects (e.g., individual partial correlations or edges) of different networks.

Subscales of the SRS-2 and SPQ served as network nodes connected by edges that corresponded to nonzero partial correlations among symptom subscales after conditioning over all other nodes. In other words, the presence of edges connecting two nodes indicates that the nodes are “conditionally dependent” given all other nodes in the network. Green (versus orange) edges in the network indicate positive (versus negative) partial correlations, and thicker edges indicate a stronger partial correlation (Epskamp et al., 2012). The current network estimation specified a Bayes Factor of 3, which is associated with positive evidence for an effect (Kass & Raftery, 1995).

Network Centrality Metrics

Centrality metrics provide insight into how nodes are interconnected within the full network. To identify nodes that are highly connected to other nodes, we estimated expected influence (Robinaugh et al., 2016), which is defined as the sum of all edges that extend from a given node. EI was chosen over other measures of centrality given its consideration of both positive and negative edges and superior performance compared to other centrality measures (McNally, 2016; Robinaugh et al., 2016). EI values were estimated with the networktools package (Jones, 2017) in R. To characterize connectivity between “communities” of nodes within a network, we also estimated bridge strength, which is defined as the sum of the absolute value of all edges that exist between a node and all nodes that are not in the same community of that node. Here, we specified ASD and SCZ communities, which consisted of nodes corresponding to the SRS-2 and SPQ subscales, respectively. This allowed us to identify how symptoms relevant to ASD are connected to those relevant to SCZ. All centrality metrics were estimated within a Bayesian framework, yielding EI and bridge strength posterior distributions for each node instead of point estimates.

Network Comparison

Network differences were examined with respect to overall network structure and pairwise comparisons between individual edges. Overall network structure was compared by estimating the partial correlation matrix distance (CMD) (Herdin et al., 2005) to assess the degree of difference between the two symptom networks. First, the posterior predictive distribution is computed under the assumption of group equality, which provides the error that would be expected under the null model (i.e., equivalence of partial correlation matrices). Then, the CMD is estimated for the observed groups and compared to the posterior predictive distribution, from which a posterior predictive p-value is computed. Rejection of the null model indicates that the assumption of group equality is not tenable. Pairwise differences in partial correlations for each edge in the ASD and SCZ networks were estimated using a similar method by comparing posterior distributions of each edge in the networks.

Results

The mean age of participants in the ASD and SCZ groups was 24.96 years (SD = 5.77) and 25.77 years (SD = 6.65) respectively. Mean IQ was 105.36 (SD = 16.07) for ASD and 97.18 (SD = 10.41) for SCZ. Participants were matched on mean age, but there were statistically significant group differences in IQ (SCZ < ASD, t(131.69) = 79.16, p < 0.001) and sex (χ2(2) = 10.04, p < 0.01) with 77% males in the ASD group and 82% males in the SCZ group.

ASD Network

The symptom network for participants with a confirmed ASD diagnosis is shown in Fig. 1A. In the ASD network (Fig. 2A), restricted and repetitive behaviors (RRBs) [posterior mean (sd) = 1.71 (0.31), 95% credible interval = (1.12, 2.35)] and social communication [posterior mean (sd) = 1.45 (0.27), 95% credible interval = (1.11, 2.16)] had the highest expected influence (EI). Furthermore, all five of the SRS-2 nodes [restricted and repetitive behaviors (2.00), social motivation (1.97), social communication (1.66), social awareness (1.66), and social cognition (1.65)] comprised the top five bridge strength estimates (Fig. 3A), suggesting that they were the most connected to SPQ nodes. Specifically, having more RRBs was linked to eccentric behavior and less social motivation was linked to social anxiety.

Fig. 1.

Fig. 1

Open in a new tab

Partial correlation networks of autism and schizophrenia symptoms in adults with confirmed diagnoses of autism (ASD) and schizophrenia (SCZ). Side-by-side partial correlation networks of SRS-2 and SPQ subscales as nodes for participants with A confirmed ASD and B confirmed SCZ diagnoses. Each edge (partial correlation between two nodes) is denoted by a weight represented by the thickness of the line. SRS_Aw SRS-2 Social awareness subscale, SRS_SCg SRS-2 Social cognition subscale, SRS_SCm SRS-2 Social communication subscale, SRS_SM SRS-2 Social motivation subscale, SRS_RRB SRS-2 Restricted and repetitive behaviors subscale, SPQ_IR Ideas of reference, SPQ_SAx Social anxiety, SPQ_OB Odd Beliefs, SPQ_UP Unusual perceptions, SPQ_Ec Eccentric behavior, SPQ_NF No Close friends, SPQ_OS Odd speech, SPQ_CA Constricted affect, SPQ_Su Suspiciousness

Fig. 2.

Fig. 2

Open in a new tab

Posterior distributions of expected influence for each node in the A autism and B schizophrenia networks. The Bayesian network analysis approach yields posterior distributions to estimate expected influence for each node in the network. Greater expected influence (distributions shifted to the right) indicates that the node is more central to the network. SRS_Aw SRS-2 Social awareness subscale, SRS_SCg SRS-2 Social cognition subscale, SRS_SCm SRS-2 Social communication subscale, SRS_SM SRS-2 Social motivation subscale, SRS_RRB SRS-2 Restricted and repetitive behaviors subscale, SPQ_IR Ideas of reference, SPQ_SAx Social anxiety, SPQ_OB Odd beliefs, SPQ_UP Unusual perceptions, SPQ_Ec Eccentric behavior, SPQ_NF No Close friends, SPQ_OS Odd speech, SPQ_CA Constricted affect, SPQ_Su Suspiciousness

Fig. 3.

Fig. 3

Open in a new tab

Posterior distributions of bridge strength for each node in the A autism and B schizophrenia networks. The Bayesian network analysis approach yields posterior distributions to estimate bridge strength for each node in the network. Bridge strength corresponds to the degree to which autism-relevant clinical features (SRS-2 subscales) are connected to schizophrenia-relevant features (SPQ subscales). SRS_Aw SRS-2 Social awareness subscale, SRS_SCg SRS-2 Social cognition subscale, SRS_SCm SRS-2 Social communication subscale, SRS_SM SRS-2 Social motivation subscale, SRS_RRB SRS-2 Restricted and repetitive behaviors subscale, SPQ_IR Ideas of reference, SPQ_SAx Social anxiety, SPQ_OB Odd beliefs, SPQ_UP Unusual perceptions, SPQ_Ec Eccentric behavior, SPQ_NF No Close friends, SPQ_OS Odd speech, SPQ_CA Constricted affect, SPQ_Su Suspiciousness

SCZ Network

The symptom network for participants with a confirmed SCZ diagnosis is shown in Fig. 1B. As shown in Fig. 2B, social communication [posterior mean (sd) = 1.95 (0.52), 95% credible interval = (1.02, 3.04)] had the highest EI, followed by ideas of reference [posterior mean (sd) = 1.70 (0.52), 95% credible interval = (0.74, 2.75)] and eccentric behavior [posterior mean (sd) = 1.25 (0.46), 95% credible interval = (0.47, 2.22)]. Bridge strength (Fig. 3B) was highest for social communication (2.73), social awareness (2.69), social motivation (2.64), and RRBs (2.46), indicating that these ASD symptoms were connected to SCZ symptoms. Greater impairments in social motivation and social cognition were associated with lower scores on suspiciousness and having no close friends, respectively. Presence of RRBs was negatively associated with social anxiety.

Comparison of ASD and SCZ Symptom Networks

Comparison of overall symptom networks indicated that the ASD and SCZ networks are statistically different with a correlation matrix distance value of 0.77, p = 0.001. Pairwise differences in partial correlations for each edge in the ASD and SCZ networks were estimated based on a difference between posterior distributions and are depicted in Fig. 4. We were particularly interested in differences in edges connecting SRS and SPQ nodes. Significant between-network differences in partial correlations were identified and corresponded to edges connecting SRS-2 social motivation to SPQ suspiciousness and SRS-2 RRBs to SPQ social anxiety and SPQ unusual perceptions.

Fig. 4.

Fig. 4

Open in a new tab

Pairwise differences for each possible edge or relation in the autism and schizophrenia networks. Posterior mean estimates of the difference between each edge in the ASD and SCZ networks. SRS_Aw SRS-2 Social awareness subscale, SRS_SCg SRS-2 Social cognition subscale, SRS_SCm SRS-2 Social communication subscale, SRS_SM SRS-2 Social motivation subscale, SRS_RRB SRS-2 Restricted and repetitive behaviors subscale, SPQ_IR Ideas of reference, SPQ_SAx Social anxiety, SPQ_OB Odd beliefs, SPQ_UP Unusual perceptions, SPQ_Ec Eccentric behavior, SPQ_NF No Close friends, SPQ_OS Odd speech, SPQ_CA Constricted affect, SPQ_Su Suspiciousness

The difference in means of the posterior distributions of estimated edge weights (ASD network edge – SCZ network edge) connecting social motivation to suspiciousness was rASD-SCZ = 0.55 [sd = 0.21; 95% credible interval = (0.10, 0.92)]. The difference in edge weights connecting RRBs to social anxiety was rASD-SCZ = 0.55 [sd = 0.20; 95% credible interval = (0.12, 0.91)]. The difference in edge weights connecting social communication to eccentric behavior was rASD-SCZ = 0.50 [sd = 0.22; 95% credible interval = (0.05, 0.91)]. Lastly, the difference in edge weights connecting social motivation to constricted affect was rASD-SCZ = 0.49 [sd = 0.22; 95% credible interval = (0.04, 0.88)].

Further examination showed that the edge weight connecting social motivation and suspiciousness was significantly negative in the SCZ network [rSCZ = − 0.42, 95% credible interval = (− 0.75, − 0.004)], whereas it was not significantly different from zero in the ASD network [rASD = 0.15, 95% credible interval = (− 0.19, 0.45)]. This pattern of effects was also seen for the edge weight connecting RRBs to social anxiety [rSCZ = − 0.47, 95% credible interval = (− 0.76, − 0.01) vs. rASD = − 0.004, 95% credible interval = (− 0.32, 0.31)]. Finally, the edges connecting social communication and eccentric behavior [rSCZ = − 0.21, 95% credible interval = (− 0.67, 0.32); rASD = 0.01, 95% credible interval = (− 0.35, 0.36)] and social motivation and constricted affect [rSCZ = − 0.23, 95% credible interval = (− 0.66, 0.27); rASD = 0.12, 95% credible interval = (− 0.21, 0.44)] differed based on a comparison of the posterior distributions, but each of the edges were not significantly different from zero based on their credible intervals. Considering all pairwise differences, significant between-network (i.e., between-group) differences were due to negative symptom relations that were present in the SCZ network that were not present in the ASD network.

Discussion

This study applied network analysis to a clinical sample of individuals meeting clinician-confirmed diagnostic criteria for ASD and SCZ. Comparison of symptom networks with autism- and schizophrenia-relevant symptoms revealed distinct network structures of ASD and SCZ symptoms (i.e., differences in overall symptom organization) for individuals diagnosed with ASD compared to SCZ. Interestingly, the node corresponding to ASD-related difficulties in social communication was the most central node for both networks, while ASD-related RRBs and SCZ-related cognitive and perceptual symptoms were most central to the ASD and SCZ groups, respectively. By examining both overall network structure and individual node importance via estimates of node centrality, results clarify points of convergence and divergence between the two conditions and help to resolve the contradictory ideas that ASD and SCZ are distinct clinical conditions yet share common symptomatology. Specifically, the conditions are characterized by shared symptoms, but symptom organization differentiates the two conditions.

The most influential nodes in the symptom network for individuals with confirmed ASD diagnoses corresponded to the RRBs and social communication subscales of the SRS-2, which is expected and consistent with well-established symptom criteria for ASD. As assessed by the SRS-2, social communication refers to one’s ability to engage in reciprocal social interaction, communicate emotions to others, utilize appropriate eye contact when interacting, and maintain social connections. Notably, social communication was also identified as having the highest expected influence in the symptom network for individuals with confirmed SCZ diagnoses, followed by ideas of reference and eccentric behavior. Ideas of reference refer to when an individual attributes strong personal significance to random or coincidental events and eccentric behavior may include odd or unusual habits, mannerisms, and behaviors. Results are consistent with previous work emphasizing persistent social difficulties in individuals with SCZ (Couture et al., 2010; Dickinson et al., 2007; Fernandes et al., 2018; Hooley, 2010; Oliver et al., 2020; Pinkham et al., 2020; Sasson et al., 2016; St Pourcain et al., 2018; Tordjman et al., 2019) and highlight overlap in observable social symptomatology in ASD and SCZ. Indeed, on laboratory-based tasks assessing facets of social cognition, both populations exhibit difficulties in emotion recognition, social perception, and theory of mind compared to non-clinical samples (Pinkham et al., 2020), as well as aberrations in social perceptual processes and reduced neural activation of the temporal-parietal occipital junction, superior temporal sulcus, and other brain regions relevant to perceiving and responding to social stimuli compared to healthy controls (Pinkham et al., 2008; Wible, 2012). Though studies indicate both common and distinct neural origins underlying social cognitive processes (Chen et al., 2017; Eack et al., 2017; Martínez et al., 2019; Mastrovito et al., 2018), the similar behavioral manifestations of social difficulties in ASD and SCZ suggest that targeting social cognition and communication would be of particular relevance for symptom improvement in both populations. Given that social communication had the highest node centrality for both networks, targeting this node could change the overall symptom networks (i.e., the manifestation of the clinical conditions) and subsequently reduce clinical severity and functional impairment.

Between-network comparison of node expected influence indicates that the conditions are best differentiated by ASD-specific restricted interests and repetitive behaviors and SCZ-specific cognitive-perceptual (ideas of reference), and disorganized (eccentric or odd behavior) symptoms. This is consistent with recent commentary and empirical work supporting the applicability of the positive, negative, and cognitive categorization of symptoms from the SCZ literature to ASD, in which positive features of ASD include the presence of atypical behaviors (e.g., stereotypic motor behaviors, echolalia, circumscribed interests), negative features are reflected by the absence of expected behaviors (e.g., reduced eye contact, facial expressions, and social engagement), and cognitive features include cognitive rigidity, uneven executive functioning profiles, and impaired theory of mind (Foss-Feig et al., 2016; Trevisan et al., 2020). As social difficulties measured by the SRS-2 primarily reflect the absence of expected social functions, shared node centrality of social communication symptoms indicates that the two conditions may be best differentiated by “positive symptoms” (e.g., RRBs) rather than “negative symptoms” (i.e., social withdrawal, flat affect).

Specific examination of individual edge weights that differ between the two networks offers insight into how symptom interrelations differ in ASD and SCZ. In the SCZ network, reduced social motivation was negatively associated with suspiciousness, whereas this relation was not identified as a significant edge in the ASD network. This finding is consistent with the social deafferentation hypothesis of SCZ (Hoffman, 2007), which posits that high levels of social withdrawal and isolation may prompt a diathesis for maladaptive social cognition in the form of hallucinations and delusions where individuals attribute inaccurate social and emotionally compelling information to other persons, agents, or environmental stimuli. In this framework individuals who derive reward from social interaction (i.e., those who are socially motivated) may be particularly vulnerable to the onset of positive SCZ symptoms. Importantly, this differs from the social motivation hypothesis of ASD (Dawson et al., 2002, 2005), which suggests that social difficulties attributed to core features of ASD are the consequence of reduced motivation to engage with social stimuli (e.g., attention to faces) starting in early development. This suggests developmental differences in the role of social motivation in ASD and SCZ: whereas diminished social motivation in early developmental periods may explain subsequent difficulties for autistic individuals to attend to and engage with social stimuli, social disconnection paired with intact social motivation later in life may contribute to the onset of positive cognitive and perceptual symptoms seen in SCZ.

The role of social motivation was also highlighted in estimates of bridge strength, which help identify which symptoms relevant to ASD and SCZ are implicated in connecting or “bridging” the two conditions. In the ASD network, RRBs and social motivation were the strongest bridge symptoms, while social communication, social awareness, and social motivation were the strongest in the SCZ network. Though beyond the scope of the current cross-sectional study, bridge strength results suggest that aberrations in social perceptual and motivational processes (e.g., social anhedonia) are trans-diagnostically relevant and explain symptom overlap and high rates of co-occurrence of ASD and SCZ. Given that the ASD and SCZ networks in the current study included individuals with either one diagnosis or the other, future work investigating the symptom networks of individuals with both clinical conditions would be helpful to further probe bridge symptoms underlying comorbidity. Considering topographical symptom overlap in ASD and SCZ, results also suggest that instruments relying on a symptom checklist approach would benefit from additional probes assessing the perceptions and motivations of the individual and subsequent functions of clinical symptoms. For example, whether a person disengages from social stimuli in favor of attending to a circumscribed interest or to engage with delusional cognitions may inform diagnostic decision making and subsequent treatment planning.

Results of this study offer recommendations for clinical application with respect to both treatment and assessment of ASD and SCZ. Shared centrality of social communication symptoms in the symptom networks for participants with confirmed diagnoses of ASD and SCZ suggests that the cluster of interrelated symptoms (networks) that are used to define the two conditions are predominantly “held together” by social communication. Intervening on this node could change the overall network structures and subsequent phenotypic presentations. Results suggest that clinical care for both populations should prioritize, monitor, and target social functioning. Group-based social skills interventions focused on learning and practicing social skills and increasing social opportunities for autistic adults may help to increase social understanding, improve social functioning, reduce loneliness, and potentially reduce co-occurring psychiatric symptoms in autistic adults (Spain & Blainey, 2015). Social skills interventions have also shown promise for improving social knowledge and functioning in schizophrenia and reducing negative symptoms (e.g., flat affect, amotivation) (Kurtz & Mueser, 2008). Results from this study support continued efforts to develop interventions for improving social skills and competence in both ASD and SCZ. Given that social symptoms were also especially important in bridging autism- and schizophrenia-related symptoms in the SCZ network, such interventions may reduce onset of psychosis or other schizophrenia symptoms in autistic adults.

Results also offer suggestions for improving differential diagnosis and emphasize the clinical utility of assessing presence of restricted interests and repetitive behaviors and/or positive symptoms of SCZ (e.g., delusions, hallucinations) to distinguish ASD from SCZ. In clinical practice, this distinction is often obfuscated by topographic symptom overlap (Chandrasekhar et al., 2020). For example, ASD-related rigid and perseverative thinking may resemble paranoid or delusional thinking. Sensory processing differences in ASD (e.g., preoccupation with internal or external sensory stimuli) or marked social withdrawal may resemble negative symptoms of SCZ, such as reduced motivation, decreased social engagement, and anhedonia. In such cases, it is important to assess the developmental time course of symptoms to determine whether the current presentation is different from the individual’s baseline. In addition, collecting collateral information from multiple sources (e.g., parents, caregivers, and teachers) will be helpful to assess temporality and consistency of symptoms to guide whether there is true co-occurrence of schizophrenia symptoms or if persevera-tive cognitions, idiosyncratic behaviors, and reduced social engagement are best attributed to ASD.

Though this study was novel in its application of network analysis to a sample of individuals with clinician-confirmed diagnoses, it is not without limitations. A primary limitation of the current study is the limited sample size, which hinders robustness and replicability of results in the network analysis framework. The small sample size and exploratory nature also informed our decision not to include IQ and sex as nodes in the symptom networks despite identification of between-group differences. Adoption of a Bayesian approach to network analysis mitigated the sample size concern and made credible our interpretations given available data. Considering the relative scarcity of cross-characterized clinical samples of individuals with ASD and SCZ (i.e., characterized on measures of both conditions) in clinical research compared to other more common psychopathologies (e.g., depression and anxiety), this sample is of unique value; these results emphasize the importance of this ambitious research approach and mark larger cross-characterized samples as an important target for future research. As an initial application of network analysis to examine symptom overlap and interrelations in ASD and SCZ, we specified inventory subscales as nodes of interest. As such, node relations offer insight into broad symptom domains but do not directly probe specific mechanisms (e.g., cognitive, social perceptual, or reward processes). Additionally, some literature cautions the use of self-report measures in clinical populations with varying levels of social and emotional insight (Mazefsky et al., 2011). We utilized measures that have previously been validated as self-report measures for adults with ASD and SCZ without intellectual disability but note that multiple informants and incorporation of behavioral or neuropsychological measures in future studies would further corroborate and strengthen clinical characterization.

The current study encourages future work using larger samples and item-level data is needed to examine replicability of results to include demographic factors (e.g., leveraging moderated graphical models to examine differential network structures based on sex) and provide more granular detail into the similarities and differences in more specific facets of social communication and functioning between the two conditions. Longitudinal studies will allow for the modeling of directed networks to examine not just node relations, but causal effects from one symptom to another. Considering that co-occurring conditions are well-documented in ASD, we also encourage future work using larger samples sizes and network analytic perspectives to understand how ASD symptoms may relate to features of other common co-occurring conditions, such as anxiety, depression, and communication disorders.

Using network analysis, the current study identified a significant difference in overall network structure of relevant symptomatology for individuals with clinician-confirmed diagnoses of ASD and SCZ while highlighting common social communication difficulties for individuals affected by both conditions. This mirrors real-world clinical contexts when differential diagnosis is complicated by both high prevalence of comorbidity and presence of symptom overlap. In such cases, clinicians must prioritize a functional assessment of behavior and attend to temporality, motivations, and context of symptom emergence, not just presence or absence of symptoms, to make accurate diagnoses.

Acknowledgements

We thank the participants for partnering with us in research. We also thank the clinicians who assisted in clinical assessment of the participants in the study, including Julie Wolf, Brianna Lewis, Kimberly Ellison, and Ela Jarzabek.

Funding

This work received support from National Institute of Mental Health (Grant Nos. R01 MH107426, R01 MH119172, U19 MH108206), and Hilibrand Foundation.

Conflict of interest

James C. McPartland consults with Customer Value Partners, Bridgebio, Determined Health, and Black Thorn Therapeutics, has received research funding from Janssen Research and Development, serves on the Scientific Advisory Boards of Pastorus and Modern Clinics, and receives royalties from Guilford Press, Lambert, and Springer.

References

  1. Borsboom D, & Cramer AO (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91–121. [DOI] [PubMed] [Google Scholar]
  2. Chandrasekhar T, Copeland JN, Spanos M, & Sikich L (2020). Autism, psychosis, or both? Unraveling complex patient presentations. Child and Adolescent Psychiatric Clinics, 29(1), 103–113. [DOI] [PubMed] [Google Scholar]
  3. Chen H, Uddin LQ, Duan X, Zheng J, Long Z, Zhang Y, Guo X, Zhang Y, Zhao J, & Chen H (2017). Shared atypical default mode and salience network functional connectivity between autism and schizophrenia. Autism Research, 10(11), 1776–1786. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Chien YL, Wu CS, & Tsai HJ (2021). The comorbidity of schizophrenia spectrum and mood disorders in autism spectrum disorder. Autism Research, 14(3), 571–581. [DOI] [PubMed] [Google Scholar]
  5. Constantino JN, & Gruber CP (2012). Social responsiveness scale: SRS-2. Western Psychological Services Torrance. [Google Scholar]
  6. Couture S, Penn D, Losh M, Adolphs R, Hurley R, & Piven J (2010). Comparison of social cognitive functioning in schizophrenia and high functioning autism: More convergence than divergence. Psychological Medicine, 40(4), 569–579. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Dawson G, Carver L, Meltzoff AN, Panagiotides H, McPartland J, & Webb SJ (2002). Neural correlates of face and object recognition in young children with autism spectrum disorder, developmental delay, and typical development. Child Development, 73(3), 700–717. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Dawson G, Webb SJ, & McPartland J (2005). Understanding the nature of face processing impairment in autism: Insights from behavioral and electrophysiological studies. Developmental Neuropsychology, 27(3), 403–424. [DOI] [PubMed] [Google Scholar]
  9. De Crescenzo F, Postorino V, Siracusano M, Riccioni A, Armando M, Curatolo P, & Mazzone L (2019). Autistic symptoms in schizophrenia spectrum disorders: A systematic review and meta-analysis. Frontiers in Psychiatry, 10, 78. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Dickinson D, Bellack AS, & Gold JM (2007). Social/communication skills, cognition, and vocational functioning in schizophrenia. Schizophrenia Bulletin, 33(5), 1213–1220. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Eack SM, Wojtalik JA, Keshavan MS, & Minshew NJ (2017). Social-cognitive brain function and connectivity during visual perspective-taking in autism and schizophrenia. Schizophrenia Research, 183, 102–109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Epskamp S, Borsboom D, & Fried EI (2018). Estimating psychological networks and their accuracy: A tutorial paper. Behavior Research Methods, 50(1), 195–212. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Epskamp S, Cramer AO, Waldorp LJ, Schmittmann VD, & Borsboom D (2012). qgraph: Network visualizations of relationships in psychometric data. Journal of Statistical Software, 48(1), 1–18. [Google Scholar]
  14. Fernandes JM, Cajão R, Lopes R, Jerónimo R, & Barahona-Corrêa JB (2018). Social cognition in schizophrenia and autism spectrum disorders: A systematic review and meta-analysis of direct comparisons. Frontiers in Psychiatry, 9, 504. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Foss-Feig JH, McPartland JC, Anticevic A, & Wolf J (2016). Re-conceptualizing ASD within a dimensional framework: Positive, negative, and cognitive feature clusters. Journal of Autism and Developmental Disorders, 46(1), 342–351. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Herdin M, Czink N, Ozcelik H, & Bonek E (2005). Correlation matrix distance, a meaningful measure for evaluation of non-stationary MIMO channels. 2005 IEEE 61st Vehicular Technology Conference, [Google Scholar]
  17. Hoffman RE (2007). A social deafferentation hypothesis for induction of active schizophrenia. Schizophrenia Bulletin, 33(5), 1066–1070. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hooley JM (2010). Social factors in schizophrenia. Current Directions in Psychological Science, 19(4), 238–242. [Google Scholar]
  19. Insel T, Cuthbert B, Garvey M, Heinssen R, Pine DS, Quinn K , Sanislow C, & Wang P (2010). Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders. The American Psychiatric Association, 167(7), 748–751. 10.1176/appi.ajp.2010.09091379 [DOI] [PubMed] [Google Scholar]
  20. Jones P (2017). Networktools: Assorted tools for identifying important nodes in networks. R package version 1.1. 0 Computer Software. Retrieved from https://CRAN.Rproject.org/package=networktools [Google Scholar]
  21. Kass RE, & Raftery AE (1995). Bayes factors. Journal of the American Statistical Association, 90(430), 773–795. [Google Scholar]
  22. Kästner A, Begemann M, Michel TM, Everts S, Stepniak B, Bach C, Poustka L, Becker J, Banaschewski T, & Dose M (2015). Autism beyond diagnostic categories: Characterization of autistic phenotypes in schizophrenia. BMC Psychiatry, 15(1), 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kay SR, Fiszbein A, & Opler LA (1987). The Positive and Negative Syndrome Scale (PANSS) for schizophrenia. Schizophrenia Bulletin, 13(2), 261–276. 10.1093/schbul/13.2.261 [DOI] [PubMed] [Google Scholar]
  24. Kurtz MM, & Mueser KT (2008). A meta-analysis of controlled research on social skills training for schizophrenia. Journal of Consulting and Clinical Psychology, 76(3), 491. [DOI] [PubMed] [Google Scholar]
  25. Lauritzen SL, & Wermuth N (1989). Graphical models for associations between variables, some of which are qualitative and some quantitative. The Annals of Statistics, 17(1), 31–57. 10.1214/aos/1176347003 [DOI] [Google Scholar]
  26. Lord C, Rutter M, DiLavore P, Risi S, Gotham K, & Bishop S (2012). Autism diagnostic observation schedule,(ADOS-2) modules 1–4. Western Psychological Services. [Google Scholar]
  27. Martínez A, Tobe R, Dias EC, Ardekani BA, Veenstra-Vander-Weele J, Patel G, Breland M, Lieval A, Silipo G, & Javitt DC (2019). Differential patterns of visual sensory alteration underlying face emotion recognition impairment and motion perception deficits in schizophrenia and autism spectrum disorder. Biological Psychiatry, 86(7), 557–567. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Mastrovito D, Hanson C, & Hanson SJ (2018). Differences in atypical resting-state effective connectivity distinguish autism from schizophrenia. NeuroImage: Clinical, 18, 367–376. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Mazefsky C, Kao J, & Oswald D (2011). Preliminary evidence suggesting caution in the use of psychiatric self-report measures with adolescents with high-functioning autism spectrum disorders. Research in Autism Spectrum Disorders, 5(1), 164–174. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. McNally RJ (2016). Can network analysis transform psychopathology? Behaviour Research and Therapy, 86, 95–104. [DOI] [PubMed] [Google Scholar]
  31. McNally RJ (2021). Network analysis of psychopathology: Controversies and challenges. Annual Review of Clinical Psychology, 17, 31–53. [DOI] [PubMed] [Google Scholar]
  32. Oliver LD, Moxon-Emre I, Lai M-C, Grennan L, Voineskos AN, & Ameis SH (2021). Social cognitive performance in schizophrenia spectrum disorders compared with autism spectrum disorder: A systematic review, meta-analysis, and meta-regression. JAMA Psychiatry, 78(3), 281–292. 10.1001/jamapsychiatry.2020.3908 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Pinkham AE, Hopfinger JB, Pelphrey KA, Piven J, & Penn DL (2008). Neural bases for impaired social cognition in schizophrenia and autism spectrum disorders. Schizophrenia Research, 99(1–3), 164–175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Pinkham AE, Morrison KE, Penn DL, Harvey PD, Kelsven S, Ludwig K, & Sasson NJ (2020). Comprehensive comparison of social cognitive performance in autism spectrum disorder and schizophrenia. Psychological Medicine, 50(15), 2557–2565. [DOI] [PubMed] [Google Scholar]
  35. R Core Team. (2020). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. [Google Scholar]
  36. Raine A (1991). The SPQ: a scale for the assessment of schizotypal personality based on DSM-III-R criteria. Schizophrenia Bulletin, 17(4), 555–564. [DOI] [PubMed] [Google Scholar]
  37. Robinaugh DJ, Millner AJ, & McNally RJ (2016). Identifying highly influential nodes in the complicated grief network. Journal of Abnormal Psychology, 125(6), 747. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Sasson NJ, Pinkham AE, Weittenhiller LP, Faso DJ, & Simpson C (2016). Context effects on facial affect recognition in schizophrenia and autism: Behavioral and eye-tracking evidence. Schizophrenia Bulletin, 42(3), 675–683. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Sobel ME (1997). Measurement, causation and local independence in latent variable models. Latent variable modeling and applications to causality (pp. 11–28). Springer. [Google Scholar]
  40. Spain D, & Blainey SH (2015). Group social skills interventions for adults with high-functioning autism spectrum disorders: A systematic review. Autism, 19(7), 874–886. [DOI] [PubMed] [Google Scholar]
  41. St Pourcain B, Robinson EB, Anttila V, Sullivan BB, Maller J, Golding J, Skuse D, Ring S, Evans DM, & Zammit S (2018). ASD and schizophrenia show distinct developmental profiles in common genetic overlap with population-based social communication difficulties. Molecular Psychiatry, 23(2), 263–270. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Tordjman S, Celume M, Denis L, Motillon T, & Keromnes G (2019). Reframing schizophrenia and autism as bodily self-consciousness disorders leading to a deficit of theory of mind and empathy with social communication impairments. Neuroscience & Biobehavioral Reviews, 103, 401–413. [DOI] [PubMed] [Google Scholar]
  43. Trevisan DA, Foss-Feig JH, Naples AJ, Srihari V, Anticevic A, & McPartland JC (2020). Autism spectrum disorder and schizophrenia are better differentiated by positive symptoms than negative symptoms. Frontiers in Psychiatry, 11, 548. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Wible CG (2012). Schizophrenia as a disorder of social communication. Schizophrenia Research and Treatment. 10.1155/2012/920485 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Williams DR, & Mulder J (2019). Bggm: AR package for bayesian gaussian graphical models.
  46. Williams DR, & Mulder J (2020). BGGM: Bayesian Gaussian graphical models in R. Journal of Open Source Software, 5(51), 2111. [Google Scholar]
  47. Williams DR, Rast P, Pericchi LR, & Mulder J (2020). Comparing Gaussian graphical models with the posterior predictive distribution and Bayesian model selection. Psychological Methods, 25(5), 653. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Zhou H-Y, Yang H-X, Gong J-B, Cheung EF, Gooding DC, Park S, & Chan RC (2019). Revisiting the overlap between autistic and schizotypal traits in the non-clinical population using meta-analysis and network analysis. Schizophrenia Research, 212, 6–14. [DOI] [PubMed] [Google Scholar]

RESOURCES