Multimodal Machine Learning Prediction of Comorbid Depression in Recent-Onset Psychosis: Evidence from the PRONIA Study

This project aims to identify and validate transdiagnostic predictors of depressive comorbidity in psychotic disorders across different illness stages using multimodal machine learning approaches. Building on previously work (https://doi.org/10.1017/S0033291725100950) and developed models in first-episode psychosis (FEP) cohorts (EUFEST and RAISE-ETP), the present study extends this work to the PRONIA study, with a particular focus on individuals with recent onset psychosis (ROP). The central objective is to assess the transportability and generalizability of predictive models of depression symptomatology across overlapping diagnostic categories (psychosis, depression, and clinical high risk states) and across stages of illness progression. Using a stratified multimodal ensembling framework, we will integrate clinical, cognitive, neuroimaging, and biological data to predict future depression comorbidities in the recent onset of psychosis (ROP) group in the PRONIA study. Models will be trained within the PRONIA ROP cohort and externally validated across multiple independent samples with the data modalities available across cohorts (clinical, quality of life, blood samples), including FEP cohorts (EUFEST and RAISE-ETP). We will also validate across PRONIA groups of recent onset depression (ROD), and clinical high-risk (CHR) groups. This design allows for a systematic evaluation of whether predictive signatures of depression are conserved across diagnostic boundaries and illness stages, thereby informing transdiagnostic and staging-oriented models of psychopathology.