Background: Identifying neuroanatomical correlates of clinical prediction models may offer pathophysiological insights into the clinical high-risk states for psychosis (CHR-P) and unveil new therapeutic targets for early intervention.

Study design: We used the North American Prodrome Longitudinal Study (NAPLS-2) risk calculator to obtain psychosis risk scores for 315 CHR-P (M = 23.85, SD = ± 5.64; female: 164) and 295 recent-onset depression (M = 25.11, SD = ± 6.21; female: 144) patients from the Personalized Prognostic Tools for Early Psychosis Management (PRONIA) cohort. Voxel-based morphometry was employed to examine associations between risk scores, gray matter volume (GMV), and white matter volume (WMV). Post-hoc, we used eigenvariate extraction to explore network-level alterations associated with significant regions. Moderation analyses were conducted to understand the influence of individual NAPLS-2 risk variables on these networks (False Discovery Rate-corrected).

Study results: Reduced hippocampal GMV (${k}_E$ = 847 voxels) and cerebellar WMV (${k}_E$ = 10 423 voxels) were associated with higher risk scores. Post-hoc analyses revealed parallel structural alterations between these regions and the entorhinal cortex, anterior cingulate cortex, thalamus, anterior limb of the internal capsule, and pons. Moderation analyses showed that family risk (first-degree relative with psychotic disorder), verbal memory, and social functioning significantly influenced structural patterns.

Conclusions: Our results provide evidence for neuroanatomical correlates of the NAPLS-2 model, with alterations in hippocampal circuits suggesting a key prognostic role in the development of neurocognitive and psychosocial deficits across diagnostic boundaries. Future longitudinal studies incorporating multimodal imaging techniques should validate these findings as potential biomarkers for psychosis risk.