Social disconnection, encompassing both social isolation (objective lack of social ties) and loneliness (subjective feeling of insufficient social ties), is strongly linked to various health risks, including immune system dysregulation and increased mortality. While the prevalence of social disconnection in schizophrenia is unclear, it's estimated to be significantly higher than in the general population. Understanding the predictors of social isolation and loneliness in schizophrenia is crucial for developing effective interventions. Previous research in serious mental illness (SMI) and the general population suggests several potential predictors, including nonsocial cognition, social cognition, social anhedonia (lack of motivation for social engagement), social avoidance (motivation to avoid negative social situations), and depression. However, the relative importance of these variables in schizophrenia remains largely unknown, particularly concerning social anhedonia's unique contribution to social isolation and loneliness. This study aimed to address this gap by using a machine learning approach to investigate the predictors of social isolation and loneliness in three distinct groups: individuals with schizophrenia, a psychiatric comparison group with bipolar disorder (BD), and a community sample enriched for social isolation. The inclusion of BD as a comparison group is appropriate due to its episodic nature and clinical similarities to schizophrenia, yet it typically presents with less impairment in social cognition and motivation. The community sample, comprising individuals who self-identified as socially isolated and a group recruited using standard methods, allowed for comparisons across a wider range of social isolation levels, minimizing group-level confounding factors. The complexity inherent in numerous potentially interrelating variables necessitates a sophisticated analytical approach. LASSO regression, a machine learning technique, was chosen for its ability to account for interrelationships and model complexity, identify the best-fitting parsimonious model, and avoid overfitting.

Existing literature points to a strong association between social disconnection and negative health outcomes. Studies in SMI populations indicate links between social isolation and nonsocial cognition, social cognition, and social anhedonia. Research in the general population shows that nonsocial cognition, social cognition, social avoidance, depression, and social isolation are predictors of loneliness. While some studies have examined these predictors in relation to schizophrenia, the extent to which social anhedonia, in particular, explains unique variance in social isolation and loneliness is unclear. Previous data-driven analyses in schizophrenia highlight anhedonia as a central variable connecting multiple domains of social functioning; however, its unique contribution to specific components of social functioning requires further investigation. This research aims to address this knowledge gap.

This study involved 72 outpatients with schizophrenia, 48 with bipolar disorder (BD), and 151 community members (CS) enriched for social isolation. Participants were recruited from outpatient clinics and through online advertisements targeting those who self-identified as socially isolated. Clinical diagnoses were confirmed using the Structured Clinical Interview for DSM-5 (SCID-5), corroborated with medical records. All participants with schizophrenia and BD were clinically stable. The CS included both socially isolated and non-isolated individuals. Inclusion criteria for all groups were age (20-60), English fluency, and absence of neurological diseases, head injuries, substance use disorders, and current mood episodes. Clinical symptoms were assessed using standardized scales: the Expanded Brief Psychiatric Rating Scale (BPRS) for positive symptoms, Hamilton Depression Scale (HAM-D) for depression, Young Mania Rating Scale (YMRS) for mania, and Clinical Assessment Interview for Negative Symptoms (CAINS) for negative symptoms. Social isolation was measured using a composite score derived from the Lubben Social Network Scale, Social Disconnectedness Scale, and Role Functioning Scale. Loneliness was assessed using the UCLA Loneliness Scale (ULS Version 3). Social cognition was measured using a composite of mentalizing (TASIT), empathic accuracy, and facial affect identification. Nonsocial cognition was assessed using the neurocognitive composite of the Measurement and Treatment Research to Improve Cognition in Schizophrenia Consensus Cognitive Battery (MCCB). Social anhedonia was measured using a brief version of the Social Anhedonia Scale (SAS), and social avoidance motivation using the Sensitivity to Rejection Scale. Data analysis used LASSO regression, a machine learning technique that accounts for interrelationships among variables and prevents overfitting. Within-group and across-group LASSO regression models were fitted to predict social isolation and loneliness from the selected predictor variables. Five-fold cross-validation was used to enhance model robustness.

The three samples exhibited comparable levels of social isolation. Within-group LASSO regression models revealed that social anhedonia uniquely explained variance in social isolation and loneliness in all three samples. In schizophrenia, nonsocial cognition also uniquely predicted social isolation. Across-group analyses revealed that social anhedonia, loneliness, and nonsocial cognition demonstrated main effects on social isolation. A significant interaction was observed between social isolation and loneliness, with a stronger association in BD and CS compared to schizophrenia. Another interaction existed between social isolation and nonsocial cognition, with a weaker effect in BD relative to schizophrenia. Across-group models for loneliness showed main effects for all variables. Interactions were again observed between social isolation and loneliness in BD and CS, and between loneliness and depression in the CS. Follow-up analyses, including additional variables (positive symptoms, negative symptoms, mania symptoms, age, and gender), increased the R² values for social isolation and loneliness in schizophrenia, highlighting the significant influence of motivational negative symptoms on social isolation across all samples. This high correlation was not surprising due to the conceptual and measurement overlap between social isolation and motivational negative symptoms.

This study's findings highlight the central and transdiagnostic role of social anhedonia in predicting social isolation and loneliness. Social anhedonia's consistent prediction across diagnostic groups suggests it as a potential target for transdiagnostic interventions to reduce social disconnection. The stronger association between social isolation and loneliness in BD and CS compared to schizophrenia may be due to differences in emotional reactivity and cognitive accessibility of social experiences. The unique role of nonsocial cognition in predicting social isolation in schizophrenia aligns with its known association with this disorder. The interaction between loneliness and depression in the CS highlights the importance of considering mood-related factors in non-clinical samples. These findings suggest the potential for both transdiagnostic interventions targeting social anhedonia and schizophrenia-specific interventions targeting both social anhedonia and nonsocial cognition. Combined motivational interviewing and cognitive-behavioral therapy, which has shown promise in improving motivational negative symptoms, could be particularly beneficial. Future research should explore virtual reality-based interventions for loneliness and social anxiety.

This study demonstrates the significant role of social anhedonia in predicting social isolation and loneliness across schizophrenia, bipolar disorder, and a community sample. Nonsocial cognition also plays a significant role specifically in schizophrenia. These findings highlight social anhedonia as a promising target for transdiagnostic interventions. Future research should explore the causal relationships between these variables and test the effectiveness of interventions targeting social anhedonia and nonsocial cognition in larger, longitudinal studies, incorporating diverse social network analyses.

This study's cross-sectional design prevents causal inferences. The community sample was not representative of the general population. Self-report measures of social anhedonia and social avoidance might limit the accuracy of the findings. Future research could benefit from longitudinal designs, more representative samples, and objective measures of social constructs.