Importance of test–retest reliability for promoting fMRI based screening and interventions in major depressive disorder

Abstract

Proponents of personalized medicine have promoted neuroimaging in three areas of clinical application for major depression: clinical prediction, outcome evaluation, and treatment, via neurofeedback. Whereas psychometric considerations such as test–retest reliability are basic precursors to clinical adoption for most clinical instruments, we show, in this article, that basic psychometrics have not been regularly attended to in fMRI of depression. For instance, no fMRI neurofeedback study has included measures of test–retest reliability, despite the implicit assumption that brain signals are stable enough to train. We consider several factors that could be useful to aid clinical translation, including (1) attending to how the BOLD response is parameterized, (2) identifying and promoting regions or voxels with stronger psychometric properties, (3) accounting for within-individual changes (e.g., in symptomatology) across time, and (4) focusing on tasks and clinical populations that are relevant for the intended clinical application. We apply these principles to published prognostic and neurofeedback data sets. The broad implication of this work is that attention to psychometrics is important for clinical adoption of mechanistic assessment, is feasible, and may improve the underlying science.