Towards Precision Medicine: Characterizing and Predicting Antidepressant Remission Heterogeneity in Late-life Depression Using Genetic Data

Late-life depression (LLD), diagnosed as major depressive disorder in older adults (≥ 60 years), is a heterogeneous disorder characterized by a unique constellation of clinical symptoms and risk factors associated with ageing, including comorbidities such as cerebrovascular disease, which impact symptomatology and response to antidepressant medication. Towards precision medicine, this thesis traverses methodologies in pharmacogenetics to identify variants associated with, and predictive of, response in older adults, specifically among those treated with the commonly used antidepressant, venlafaxine.First, using a candidate-gene approach, we show an association of the norepinephrine transporter gene (SLC6A2) with LLD remission (OR = 1.67 [1.13, 2.42], p = 0.009), which is a validated target of venlafaxine. In addition, we identify variants in novel genes of interest, particularly in the PIEZO-type mechanosensitive ion channel component 1 gene (PEIZO1; OR = 0.33 [0.21, 0.51], p = 1.42×10^-6), which implicates cardiovascular and inflammatory processes in LLD and antidepressant response. Given our interest in predicting antidepressant response, we evaluate the predictive capacity of data-driven, machine learning models which integrate clinical and genome-wide data. Compared to baseline models using only clinical features, integrated models show decreased performance due to noise added by genetic features. Furthermore, integrated models do not achieve clinical significance (remission, AUC = 0.68, Sensitivity = 0.53, Specificity = 0.75, Accuracy = 65.0%, p = 5.08×10^-6; end-of-treatment depression severity, RMSE = 7.93, Pearson’s ρ = 0.22, p = 0.033). Our results present several implications, including (1) the need for larger clinical studies targeted towards the development of inferential and predictive pharmacogenetic models; (2) the further need for leveraging in silico databases to characterize and understand the biological relevance of associated variants; and, (3) a better understanding of the computational and statistical implications for machine learning models integrating high-dimensional genome-wide and clinical data for the prediction of complex psychiatric outcomes. Ultimately, such predictive models will aid the development of decision support tools to identify older adults at high risk for inadequate response to antidepressant treatment who may benefit from additional or alternative interventions to reduce the overall healthcare burden and improve patient quality of life.

Department dc:contributor.department

Medical Science

Year dc:date.issued

2022

Author dc:creator

• Marshe, Victoria

Advisor dc:contributor.advisor

• Mueller, Daniel

• Antidepressant response
• Genome-wide association study
• Late-life depression
• Machine Learning
• Pharmacogenetics
• Venlafaxine

Handle dc:identifier.uri

http://hdl.handle.net/1807/110883

OAI identifier oai:identifier

oai:utoronto.scholaris.ca:1807/110883