Item-level psychometric data map continuous diagnostic and symptom fields across two transdiagnostic cohorts

Background. Psychiatric diagnoses remain clinically useful, but symptom burden, impairment and comorbidity vary continuously within and across disorders. We tested whether item-level psychometric data could be used to construct a low-dimensional participant space on which diagnoses and other variables vary as smooth, interpretable gradients. Methods. We analysed two independent public transdiagnostic cohorts with structured diagnostic assessment and item-level questionnaire/interview data. Within each cohort, we used intrinsic-dimension-guided item weighting followed by robust principal components analysis to derive two psychometric coordinates without using diagnostic labels. We modelled diagnoses in both cohorts, and symptom severity, functioning and treatment burden in LA5c, as smooth functions of embedding position. Discrimination was evaluated from out-of-fold predictions, including nested comparisons of embedding-only and embedding-plus-residual models. Results. In both cohorts, participants occupied continuous two-dimensional psychometric spaces, with diagnoses distributed as broad, overlapping probability gradients. In LA5c, symptom severity, functioning and treatment burden also varied systematically across the embedding. Secondary residual-feature comparisons showed isolated unadjusted gains in LA5c but none in TCP. Conclusions. Despite differing axis orientation, both cohorts showed a similar continuous organisation of psychometric and diagnostic variation.