Statistical Modelling for relating multimodal neuroimaging to clinical outcomes in order to predict patient response to depression therapy.

Obiettivo

Every year, 1 out of 15 Europeans suffer from major depression (MDD) and MDD is the third cause of Disability-adjusted life-years. Today, the available treatments are clearly insufficient; only about 50% of MDD patients respond to drug intervention. We here posit that identification of biomarkers that can predict treatment response is needed to adapt a personalized medicine approach, and most likely this will involve not a single outcome but a combination of multimodal brain imaging outcomes, psychological, genetic, and environmental data. The complexity of such data requires a complex statistical model that currently does not exist. Thus, my aim is to develop a new flexible statistical method that can take into account heterogeneous types of data. More specifically, I will develop a fully flexible Latent Variable Model (LVM) that can deal with high dimensional measurements (e.g. images), non-Gaussian variables, and non-linear relationships. I will apply this flexible LVM on existing data from depressed and healthy individuals and later expand the application to predict treatment outcomes. The latter data are currently acquired and includes a cohort of MDD patients treated with a selective serotonin reuptake inhibitor (SSRI), followed in a longitudinal design. The chosen host institution is perfectly situated to this project, as they have an established unique database including, e.g. functional Magnetic Resonance Imaging (fMRI), high resolution Positron Emission Tomography (PET), and neuropsychological test outcomes. This research project uniquely combines advanced statistical modelling of rich data sets with the ultimate aim to establish individualized depression therapy. Moreover, it forms a foundation for a more general approach to integrate brain neurobiology in terms of imaging outcomes with other patient-specific data.