¹Jeonbuk National University, Department of Psychiatry, Medical School, Jeonju, Korea, Republic of, ²Bongseng Memorial Hospital, Department of Psychiatry, Busan, Korea, Republic of, ³Jeonbuk National University Hospital, Department of Psychiatry, Jeonju, Korea, Republic of

Background: Psychosis is increasingly conceptualized as a disorder of disrupted brain network connectivity, with altered network topology linked to symptom severity and functional decline. Traditional static connectivity measures may fail to capture the brain’s ability to adapt when its network is perturbed, which may be critical for understanding illness progression and treatment response. Brain network resilience, quantified through perturbation simulations, has emerged as a promising biomarker for staging and prognosis.

Aims & Objectives: This study aimed to evaluate brain network resilience as a dynamic biomarker for staging and prognosis across the psychosis spectrum by characterizing stage-related alterations in resilience metrics, integrating resilience-based markers into a data-driven Subtype and Stage Inference (SuStaIn) model, and examining associations between SuStaIn-derived stages, cognitive performance, symptom severity, and relapse risk.

Method: 

1. Recovered psychosis study: We investigated 30 patients who had recovered from psychosis following antipsychotic discontinuation and compared them with 50 healthy participants. Using resting-state functional magnetic resonance imaging and graph-based analyses, network robustness was assessed under random and targeted attack simulations. Disintegration points were defined as the thresholds at which the giant connected component collapsed.
2. First-episode psychosis study: Data from 340 patients with first-episode schizophrenia spectrum disorders and 420 healthy participants from two independent sites were analyzed. K-shell decomposition was applied to derive resilience measures, including k-core resilience, maximum core resilience, and k-shell resilience. Associations with cognitive performance and clinical symptom severity were examined.
3. Staging and prognosis study: These approaches are being extended to a large multi-site cohort of more than 2,000 participants from Korea and China. Resting-state functional magnetic resonance imaging data are harmonized using ComBat and converted into thresholded functional connectivity matrices using orthogonal minimum spanning tree and proportional thresholding. Derived resilience metrics are incorporated into the Subtype and Stage Inference model to infer biologically informed stages and subtypes with probabilistic stage assignment. Associations with clinical symptoms, social and occupational functioning, cognition, and treatment response are being evaluated.

Results: 

1. Recovered psychosis study: Patients who relapsed after medication discontinuation exhibited lower global efficiency, longer characteristic path length, and reduced network robustness compared with those who maintained recovery. Disintegration points occurred at earlier attack thresholds in relapsed patients, indicating reduced resilience.
2. First-episode psychosis study: Patients with schizophrenia spectrum disorders showed significantly lower k-core and k-shell resilience compared with healthy participants. Reduced resilience was strongly associated with poorer executive functioning and greater clinical symptom severity.
3. Staging and prognosis study: Preliminary staging analyses demonstrate that patients exhibit significantly reduced resilience (p < 0.001), earlier network disintegration, and smaller maximum core sizes compared with healthy participants. These dynamic biomarkers are currently being integrated into the Subtype and Stage Inference model to evaluate their utility for biologically informed staging.

Discussion & Conclusions: Together, these studies highlight brain network resilience as a clinically relevant biomarker for psychosis staging and prognosis. Integrating resilience-based markers with Subtype and Stage Inference modeling provides a robust framework for refining disease staging, predicting relapse risk, identifying neurobiological subtypes, and evaluating treatment response. This approach may ultimately enable stage-specific and personalized interventions in psychosis.