Prediction of Mental Health crises based on Electronic Health Records - Probabilistic and Machine Learning Models for Clinical Applications

Aug 4, 2026

Upload by Rowan

This doctoral thesis investigates how Electronic Health Records combined with probabilistic and machine learning methods can support relapse-prevention for mental health patients. The research predicts relapse risk across the full interval between two consecutive mental health crises, aiming to enable timely preventive interventions and more efficient use of healthcare resources. It also proposes a personalized monitoring policy determining when patients reach stability and introduces a regularly updated risk model during that stable phase. The work addresses challenges of translating machine learning to clinical workflows and demonstrates successful implementation and cross-system transfer.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages208

Reading Time8 h 44 min

AccessFree to download

1 Introduction

1.1 Background

1.2 Main Challenges

Prediction of Mental Health crises based on Electronic Health Records: Probabilistic and Machine Learning Models for Clinical Applications Roger Garriga Calleja DOCTORAL THESIS UPF / 2023 THESIS SUPERVISORS Dr. Aleksandar Mati Prof. Gábor Lugosi Dept. of Information and Communication Technologies Per a l’Anna, per la sevapaciència i suport durant tot el doctorat, i per l’Andreu, que cada dia ens omple d’alegria a tots iii I want to thank several people who have inspired and given me support and guidance through the development of this thesis. First, I want to thank my supervisors and my tutor. Aleks, thank you for suggesting me to pursue this path and guiding me since even before starting my PhD. Your advice have been invaluable from the start, you taught me how to conduct high quality Research, and how to structure and properly convey the ideas behind our work, but also how to think and connect the pieces to find my path. You have been truly inspiring to me. Gábor, thank you for enthusiastically accept the challenges I posed, for your guidance and encouragement to do rigorous work and supporting me on the path I took. Vicenç, thank you for your kind support and openness to discuss and help whenever I needed your advice. There are many people who crossed my academic and professional path who I want to thank. Thank you Javi, for taking the first steps of this work with me. José, for your encouragement and help in the moments I needed. João, for always being approachable and offering your help or advice. Oliver H., for your trust and support, for being so approachable and provide your knowledge and wisdom to guide our work. Mohammed, for sharing the PhD hurdles. Silvina, for your encouragement and inspiration. Bartek, Sandra and Jesús for inspiring me as researchers. And Iñaki, for all the guidance you gave me since the beginning, your encouragement to pursue the PhD, your confidence, all the discussions we had, your advice, and generally for being a source of inspiration. I want to extend my thanks to the people at Koa and to our collaborators from Birmingham and Rush, especially Phil, Lou, Jon, George and Niranjan, this work would not have been possible without you. Finalment vull agraïr a la meva familia. Mare, gràcies per ser-hi sempre, per ser la persona amb qui sempre m’he pogut recolzar des de ben petit. Gràcies pel teu amor incondicional i la confiança que sempre has tingut amb mi. M’has donat les eines i el suport per poder-me desenvolupar i arribar fins aquí . Anaïs, gràcies pel teu suport iser-hi sempre que et necessito, saber que puc comptar amb tu em fa estar tranquil. Toni, Juan Ángel, Maria Àngels, Marta, gràcies per estar sempre disposats a donar un cop demà i acollir-me. Anna, gràcies per la teva dedicació, paciència i suport durant tot el doctorat. Perescoltar-me i apoiar-me en els moments que he estat més baix i instar-me a seguir. Per encoratjar-me en tot moment i donar-me canya quan ho necessito. Gràcies pel teucarinyo, per ser-hi sempre i cuidar-me. Sense tu no hagués estat possible. v Abstract This thesis delves into the transformative potential of Electronic Health Records and Machine Learning in the prevention of mental health crisis relapses. Our work focuses on the prediction of these relapses throughout the entire period between two consecutive mental health crises, with the aim of facilitating timely interventions to mitigate the risk of relapse and optimize healthcare resources. Concretely, we develop a personalized policy to determine the extent of time a patient needs to be closely monitored before reaching stability and a Machine Learning model that provides a regularly updated risk of relapse during the patient stability phase. In our studies, we address key challenges in transferring a Machine Learning algorithm to clinical practice. To this end, we demonstrate the successful implementation of our Machine Learning algorithm within clinical workflows, enabling pre

FAQs

What does the thesis focus on in relation to mental health crises?

It focuses on predicting relapses by analyzing electronic health records and leveraging probabilistic and machine learning models.

How does the proposed approach use time between crises?

It models relapse risk throughout the entire period between two consecutive mental health crises to support timely interventions.

What clinical translation results does the thesis report?

It demonstrates successful implementation of the machine learning algorithm within clinical workflows and shows the model can be transferred across different healthcare systems.

Aug 4, 2026

Upload by Rowan