• 1 download

Computational psychiatry: machine learning for clinical decision support in the treatment of major depression

  • Author / Creator
    Benoit, James RA
  • The goal of this thesis is to contribute to the fields of data-driven medicine and computational psychiatry by attempting to demonstrate the viability of machine learning for use in psychiatry, specifically in predicting treatment outcomes for major depression. This is attempted in four ways: 1. Chapter 2 is original research (and an intended article) describing a way to use ML to produce a learned classifier that takes as input patient clinical features to predict symptom remission after eight weeks of a specific antidepressant therapy. 2. Chapter 3 is original research (and an intended article) describing a way to use automated machine learning software for predicting treatment response after eight weeks of antidepressant therapy. 3. Chapter 4 is a literature review updating the reader on progress in how machine learning has been applied in the fields of psychiatry and personalized medicine. 4. Chapter 5 is a viewpoint (and intended article) suggesting changes in psychiatric prescribing practice that will occur as a result of deploying machine learning tools. Chapter 2 uses data from 11 of Pfizer’s desvenlafaxine (DVS; trade name Pristiq) clinical trials to demonstrate the construction and use of a machine learned model for predicting treatment outcomes in depression after eight weeks of treatment. Results show that using pre-treatment baseline data comprising psychiatric scales, laboratory test data, demographic information, and medication-related data, is sufficient to produce a classifier capable of predicting symptom remission, defined as a Hamilton Depression Rating Scale (HAM-D) score of ≤ 7, with 69.0% accuracy, 6.9% above chance predictions (p<0.05). Chapter 3 draws from the same dataset, using the automated machine learning software (RapidMiner) to train classifiers to predict treatment response, defined as a ≥ 50% reduction in symptoms, based on the HAM-D scale. Without including early response data, classifiers were only able to predict response at 58.90%; after including early response data, classifiers were able to predict response at 70.05% accuracy. Chapter 4 is framed as a conceptual review of machine learning in personalized medicine and psychiatry, focusing on recent applications of machine learning software to psychiatric care challenges. It covers four domains: data access, movement away from traditional statistical models, knowledge translation (KT) & commercialization of machine learning technology, and futurism. Within these domains, the chapter examines the development of Electronic Medical Records (EMR’s) as they relate to personalized medicine and the interaction of health data with developing technologies such as streaming data and data ownership, the interaction of health data and machine learning, the health implementation environment, and current mental health tools being deployed commercially. Chapter 5 is a viewpoint focusing on how we anticipate machine learning will affect clinical prescribing practice. Currently, clinical trials focus on demonstrating population-level safety and efficacy of new antidepressant drugs, but do not account for variance between individual patients. Deployment of machine learning and learned tools in the clinic will give clinicians the ability to compare the probabilities of different antidepressants being effective while minimizing side-effect profiles, on a patient-by-patient basis. This chapter focuses on the possible downstream effects of clinical machine learning tool deployment at different levels of the healthcare environment. Among these four chapters, the thesis attempts to demonstrate the viability of using machine learning for prediction of psychiatric treatment outcomes, and to articulate how the field of data-driven medicine is advancing quickly toward widespread use. This work has relevance for understanding ways in which machine learning, clinical practice, and future drug development in a transition to a future that will be characterized by a more data-driven, outcome-focused environment for individual patients.

  • Subjects / Keywords
  • Graduation date
    Fall 2019
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.