Prediction modeling in transdiagnostic risk: results from the PROCAN study

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Identifying biomarkers for serious mental illnesses (SMI) is crucial for early intervention.
• Youth at transdiagnostic risk require predictive models for timely support.
• Current diagnostic approaches may benefit from multi-modal data integration.

Purpose of the Study:

• To develop a machine learning model for predicting transition to SMI in at-risk youth.
• To identify neuroimaging, clinical, and behavioral predictors of SMI transition.
• To explore the utility of multi-modal data for risk stratification.

Main Methods:

• Recruited youth (12-25 years) at transdiagnostic risk and followed for 4 years.
• Collected multi-modal data: diffusion MRI, functional MRI, clinical scales (SOPS, K10), and cognitive/behavioral measures.
• Employed machine learning (Random Forest) and graph theory analysis on neuroimaging data.

Main Results:

• Diffusion MRI-derived nodal metrics (betweenness centrality) in specific brain regions (angular gyrus, inferior temporal gyrus, amygdala, calcarine fissure) discriminated between groups.
• Scale of Psychosis-Risk Symptoms (SOPS) and K10 Distress Scale scores were significant behavioral predictors.
• A combined model of neuroimaging and clinical data showed promise for predicting transdiagnostic risk and transition.

Conclusions:

• Multi-modal data integration, particularly neuroimaging and clinical assessments, offers a promising avenue for predicting SMI transition.
• Specific brain connectivity patterns and symptom severity scales can serve as potential biomarkers.
• This predictive model supports enhanced early intervention and prevention strategies for at-risk youth.