Stage-dependent reorganization of amyloid PET region-symptom bipartite networks in drug-naïve, amyloid-positive Alzheimer's disease.

Management of neuropsychiatric symptoms (NPS) is pivotal to care in Alzheimer's disease (AD), yet their alignment with amyloid topology may vary by stage. We examined whether the community (modular) structure of a brain-region-symptom co-occurrence network differs by clinical stage, testing the hypothesis that modular organization is more pronounced in earlier AD and becomes less segregated with increasing severity. In a cross-sectional retrospective cohort from a tertiary dementia clinic, we included 301 consecutive, drug-naïve patients with probable AD and amyloid-positive PET (18F-FC119S). Patients were stratified by Clinical Dementia Rating (CDR 0.5, n = 38; CDR 1.0, n = 107; CDR 2.0, n = 156). We built bipartite networks linking six cortical regions (bilateral frontal, temporal, parietal; PET positivity from automated SUVRs) to 12 Korean Neuropsychiatric Inventory (K-NPI) domains (presence = frequency × severity ≥ 1). Network density, community structure (Louvain modularity with 2000 within-symptom permutations), node strength/centralities, and stage-matched backbones were assessed. The pooled network was dense and non-modular. Stage-stratified analyses revealed significant modularity at CDR 0.5 only, with increasing density from CDR 0.5 to 2.0 and loss of community structure thereafter. Left temporal and left frontal cortices emerged as consistent regional hubs, while parietal contributions were minimal. On the symptom side, aggression and delusion carried the largest co-occurrence burdens, followed by disinhibition and anxiety; findings were robust across backbone and sensitivity analyses. These results support a stage-dependent reorganization from modest, symptom-specific modularity in early stage AD to diffuse, hub-centric coupling in later stages. Recognizing this transition may reconcile prior inconsistencies and inform clinical strategy: targeted, domain-focused interventions when modularity persists versus global stabilization as networks densify, with implications for enrichment and endpoint selection in NPS-focused trials.