Hierarchical Clustering of Canonical Retinal–Biopsychosocial Covariation Patterns Reveals Distinct Psychosis Subgroups: A Data-Driven Study from the UK Biobank
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Retinal alterations and related health/cognitive profiles are reported in psychosis spectrum disorders (PSD), but it is unknown whether multivariate retina–biopsychosocial covariations can identify biologically distinct subgroups.
Methods
Layer-resolved retinal thicknesses were obtained from macula-centered optical coherence tomography (OCT) scans using UK Biobank data from 476 adults (238 PSD, 238 age- and sex-matched controls). Biopsychosocial measures included sociodemographic/economic, cardiometabolic, ocular, and cognitive measures. Canonical correlation analysis (CCA) was employed to characterize patterns of covariation between retinal and biopsychosocial measures. Statistically reliable retina–biopsychosocial covariations were hierarchically clustered to derive subgroups, which were compared across various profiles using general linear models with false-discovery rate (FDR) correction.
Results
CCA revealed a multivariate retinal–biopsychosocial association pattern (r=0.468, 95% CI=0.410-0.527, p<.0001) linking lower retinal thickness, primarily outer-layers with greater socioeconomic deprivation, higher BMI, poorer visual acuity and lower cognitive performance. Hierarchical clustering of CCA projections based on this multivariate pattern identified 2 internally stable groups. Patients in cluster 1 (retina-preserved, 43.7% PSD, 81.9% controls) were cytoarchitecturally and systemically similar to controls, except for thicker nerve fiber layer (RNFL, Cohen’s d =0.32, p FDR =0.0093). Cluster 2 (retina-impaired) had significantly more participants with PSD (76% PSD) and they exhibited lower thicknesses across total macula ( d =-0.80, p FDR =5.64×10 −12 ), RNFL ( d =-0.39, p FDR =0.0009), and photoreceptor layers ( d =-0.42 to -1.06, p FDR <.0001), and were also characterized by lower socioeconomic status ( d= -0.71 to -1.35, p FDR <.0001).
Conclusions
Integrating layer-resolved OCT with biopsychosocial measures can noninvasively stratify PSD into biologically meaningful subgroups, underscoring the translational potential of retinal microstructure for precision psychiatry beyond symptom-based classifications.
