Multivariate brain-behaviour associations in psychiatric disorders

Area of Science:

• Neuroscience
• Psychiatry
• Data Science

Background:

• Understanding brain-behavior associations is crucial for psychiatric disorder diagnosis and treatment.
• Traditional univariate and single multivariate approaches have limitations in capturing complex relationships.
• Emerging doubly multivariate methods, such as Canonical Correlation Analysis (CCA) and Partial Least Squares (PLS), offer advanced tools for simultaneous analysis of brain and behavior.

Purpose of the Study:

• To provide an overview of doubly multivariate methods (CCA and PLS) for brain-behavior association studies.
• To review existing literature on these methods in psychiatric disorders.
• To discuss challenges and biases in predictive modeling from a machine learning perspective.

Main Methods:

• Systematic literature review of studies employing CCA and PLS for brain-behavior associations in psychiatric disorders.
• Analysis of 39 studies across attention deficit and hyperactive disorder (ADHD), autism spectrum disorders (ASD), major depressive disorder (MDD), psychosis spectrum disorders (PSD), and transdiagnostic (TD) groups.
• Focus on brain measures (morphology, connectivity, white matter integrity) and behavioral variables (symptoms, cognition, physical health, clinical history).

Main Results:

• Most studies (67%) utilized CCA, focusing on brain morphology, resting-state functional connectivity, or fractional anisotropy.
• Common findings across diagnoses include links between clinical/cognitive symptoms and frontal brain morphology/activity, and white matter association fibers.
• Physical health and clinical history emerged as significant, yet less investigated, behavioral predictors.

Conclusions:

• Doubly multivariate approaches effectively identify complex brain-behavior associations in psychiatric disorders.
• Frontal brain regions and white matter tracts are consistently implicated across various diagnoses.
• Studies are susceptible to bias from low sample size-to-feature ratios and in-sample testing, necessitating careful methodological considerations and validation.