Using genetic and neuroimaging correlates of early psychosis to investigate the underlying mechanisms of the disorder.

Early interventions improve outcomes for patients with psychotic disorders. Therefore predicting which patients will have poor outcomes such as development of a psychotic disorder or treatment resistance is of vital importance. Biomarkers for these outcomes have been suggested from several data modalities including neuroimaging, genetics, blood biomarkers, cognitive testing, sociodemographic information, and environment. However, each biomarker is associated with only a small increase in risk. This means that models that focus on one biomarker, or even one data modality, do not predict outcomes in psychosis very well. Combining different data modalities has been shown to increase predictive power if the data present different and complementary information.

With the aim of building prediction models using multiple data domains, large multisite studies collect different data modalities from patients in the early stages of illness. These patients are then monitored over time to see if they develop the outcome of interest. However the increase in date increases the complexity of the analysis, leading to potential biases and errors.

The aim of this PhD is to develop a framework for building large multimodal prediction models. I will first write a review of all of the factors that must be considered for multimodal analysis. Then I will use the clinical/sociodemographic dataset from the OPTIMISE study to build a prediction model for treatment response, to demonstrate the factors that need to be considered with even a single modality. Finally I will use the multimodal dataset EU-GEI to build prediction models to predict development of a psychotic disorder from the Clinical High Risk state. I will take several approaches, including a literature-driven method for selecting predictors versus a data driven method where all the data is entered into the model.