Developing prediction models for the treatment of depression: statistical and machine learning approaches to optimise antidepressant use in real world

Thesis

Major Depressive Disorder (MDD) is a mental health disorder, affecting an estimated 280 million people globally and creating a substantial economic and healthcare burden. Current therapeutic strategies, encompassing pharmacological, psychological and physical modalities— are typically prescribed based on a generalised "one-size-fits-all" approach. This conventional approach has revealed constraints regarding both efficacy and safety, thus hindering realisation of optimal patient outcomes. My thesis will focus on examining first-line pharmacological antidepressant treatments.

In the present medical landscape, there is a clear shift towards precision medicine, which aims at tailoring treatment to the individual characteristics of each patient at the right time. While randomised studies provide gold-standard evidence-based data, there is a vast and growing repository of observational data. These observational data are characterised by lower internal validity, but higher generalisability than randomised data. Two main challenges include building robust multivariable prediction models and integrating these two types of data. Multivariable prediction models use a wide array of variables to forecast individual patient outcomes with greater precision. By analysing patterns and correlations within the data, these models can predict the likelihood of various treatment responses for different patients, thereby informing more personalised care strategies. For a clinically relevant advancement, we need a reliable, evidence-based decision-support tool that is underpinned by these multivariable prediction models and that can offer treatment recommendations tailored to individual patients and clinicians.

In this DPhil thesis, I leveraged a population-based medicine framework to develop multivariable prediction models, with the aim of facilitating more personalised and effective treatments for MDD, using observational data from the QResearch primary care research registry (www.QResearch.org). My work is part of the "Personalise Antidepressant Treatment for Unipolar Depression Combining Individual Choices, Risks and Big Data (PETRUSHKA)" project which aims at integrating randomised and observational data to develop and subsequently test a precision medicine approach to the pharmacological treatment of MDD.

Authors: De Crescenzo, F

• DOI: 10.5287/ora-o1j0j1eao
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: depression; real-world; machine learning; prediction
• Subjects: Predictive analytics; Antidepressants; Machine learning; Psychiatry; Depression, Mental; Clinical pharmacology