Predicció anatòmico-clínica del risc d'evolució desfavorable en el pacient amb trastorn mental cap a la medicina personalitzada

Abstract

The three main goals in therapy for patients with mental disorders are to achieve symptomatic remission, social recovery, and reduce the risk of future relapses. Estimating the risk of these goals being achieved will therefore be of utmost importance. And while it is important to improve the symptoms of the current episode to achieve symptomatic remission, it is equally important to predict the risk of relapse, as around 30% of patients with a first episode of psychosis (FEP) will relapse within the first year, and up to 80% will do so within five years. This thesis aims to create and validate a relapse risk estimation model for patients after a first episode of psychosis using clinical and magnetic resonance imaging data from different centers in Spain. The thesis consists of two articles: 1) the first one shows the importance of controlling the potential site effect when evaluating the performance of a model and provides ways to do so, and 2) the second one explains the process for creating the model, provides an easy-to-use tool for applying it, and validates it on a sample of 227 patients with an FEP. To achieve this goal, this work addresses two methodological issues: a) how to control the potential bias of having multicenter data; and b) what are the characteristics of structural brain magnetic resonance imaging and the optimal preprocessing to achieve better predictions. To control the center effect in estimating the performance of a multicenter model, first, the importance of controlling this effect with synthetic and real data is demonstrated; then, two methods are proposed to avoid the presence of this bias, and the characteristics are detailed to choose the one that best fits the data. Additionally, an easy-to-use tool is provided to control this effect easily. Regarding achieving the appropriate characteristics of structural brain magnetic resonance imaging and optimal preprocessing for better prediction, two independent databases were used to see which characteristics help to better performance of the models by comparing different configurations. For the creation of the tool, Artificial Intelligence techniques were used on brain magnetic resonance imaging data and clinical data to detect patients with a high risk of relapse after experiencing a first episode of psychosis. To do this, a model was created using a cohort of 227 subjects who had experienced an FEP. The model was able to detect patients at high risk of relapse correctly. Statistically, the model detected that patients classified as at high risk of relapse had almost five times higher risk than those classified as at low risk of relapse. Finally, in this thesis it has been demonstrated that it is possible to estimate the relapse risk of patients with an FEP using structural brain magnetic resonance data together with clinical data, and that this approach works better than using neuroimage or clinical data separately. And in addition, several tools have been provided to facilitate the use of the different methodologies presented in the thesis.

Date of Award	1 Mar 2023
Original language	Catalan
Supervisor	Joaquim Radua Castaño (Co-director), Eduard Vieta i Pascual (Co-director) & Albert Fernandez Teruel (Director)