Making augmented human intelligence in medicine practical: A case study of treating major depressive disorder

Individualized medicine tailors diagnoses and treatment options on an individual patient basis. This is a paradigm shift from choosing a treatment based on highest reported efficacy in clinical trials, which is often not effective for all individuals. In this dissertation, we assert that treatment selection and management can be individualized when clinicians assessment of disease symptoms are augmented with a few analytically identified patient-specific measures (e.g., genomics, metabolomics) that are prognostic or predictive of treatment outcomes. Patient-derived biological, clinical and symptom measures are sufficiently complex, i.e., heterogeneous, noisy and high-dimensional. The question for research then becomes: “Which few among these large complex measures are sufficient to augment the clinician’s disease assessment and treatment logic to individualize treatment decisions?” This dissertation introduces, ALMOND — Analytics and Machine Learning Framework for Actionable Intelligence from Clinical and Omics Data. As a case study, this dissertation describes how ALMOND addresses the unmet need for individualized medicine in treating major depressive disorder — the leading cause of medical disabilities worldwide. The biggest challenge in individualizing treatment of depression is in the heterogeneity of how depressive symptoms manifest between individuals, and in their varied response to the same treatment. ALMOND comprises a systematic analytical workflow to individualize antidepressant treatment by addressing the challenge of heterogeneity of major depressive disorder. First, “right patients” are identified by stratifying patients using unsupervised learning, that serves as a foundation to associate their disease states with multiple pharmacological (drug-associated) measures. Second, “right drug” selection is shown to be feasible by demonstrating that psychiatrists’ depression severity assessments augmented with pharmacogenomic measures can accurately predict remission of depressive symptoms using supervised learning. Finally, probabilistic graphs provide early and easily interpretable prognoses at the “right time” to a psychiatrist by accounting for changes in routinely assessed depressive symptoms’ severity. By choosing antidepressants that have the highest-likelihood of the patient achieving remission, the chances of persisting depressive symptoms are reduced, which is often the leading medical conditions in those who commit suicide or develop chronic illnesses.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Athreya, Arjun Prasanna

Contributors dc:contributor

• Iyer, Ravishankar K.
• Hwu, Wen-Mei
• Robinson, Gene
• Sanders, William H.
• Roth, Dan
• Weinshilboum, Richard M.

• artificial intelligence
• machine learning
• depression
• genomics

Statement dc:rights

• Copyright 2019 Arjun Prasanna Athreya

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/105013

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/105013