Cox Regression Based Modeling of Functional Connectivity and Treatment Outcome for Relapse Prediction and Disease Subtyping in Substance Use Disorder

Abstract

Functional magnetic resonance imaging (fMRI) has become one of the most widely used noninvasive neuroimaging technique in research of cognitive neurosciences and of neural mechanisms of neuropsychiatric/neurological diseases. A primary goal of fMRI-based neuroimaging studies is to identify biomarkers for brain-behavior relationship and ultimately perform individualized treatment outcome prognosis. However, the concern of inadequate validation and the nature of small sample sizes are associated with fMRI-based neuroimaging studies, both of which hinder the translation from scientific findings to clinical practice. Therefore, the current paper presents a modeling approach to predict time-dependent prognosis with fMRI-based brain metrics and follow-up data. This prediction modeling is a combination of seed-based functional connectivity and voxel-wise Cox regression analysis with built-in nested cross-validation, which has been demonstrated to be able to provide robust and unbiased model performance estimates. Demonstrated with a cohort of treatment-seeking cocaine users from psychosocial treatment programs with 6-month follow-up, our proposed modeling method is capable of identifying brain regions and related functional circuits that are predictive of certain follow-up behavior, which could provide mechanistic understanding of neuropsychiatric/neurological disease and clearly shows neuromodulation implications and can be used for individualized prognosis and treatment protocol design.