Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Epidemiology and Biostatistics

Supervisor

Lizotte, Daniel J.

Abstract

Introduction: Statin drugs are a highly efficacious treatment for hypercholesterolemia and adherent treatment with statins reduces the risk of cardiovascular disease. Although statins are generally well tolerated, myalgia (muscle pain) is a common side effect and can lead to non-compliance with treatment. Increased systemic exposure may contribute to the development of myalgia. Drug-drug interactions inhibiting statin metabolism and impaired drug transporter function may lead to decreased statin clearance. Establishing accurate predictive models is an important step towards preventing adverse drug events by titrating statin dosing to limit systemic exposure.

Objectives: 1) To develop an algorithm to select concomitant medications for incorporation into the existing systemic exposure model and assess their predictive impact; 2) to apply nonlinear techniques to model systemic statin exposure, and assess their effectiveness and feasibility; 3) to identify novel genes and corresponding single nucleotide polymorphisms by NGS in patients whose statin plasma concentrations were under-predicted using the original systemic exposure model to guide future biological research.

Methods: Data from a previously-collected prospective cohort of 130 patients prescribed rosuvastatin and 128 patients prescribed atorvastatin were used in this analysis. Concomitant medications were selected using penalized. Stable feature selection was achieved by repeated cross validation. Generalized additive models (GAMs) and support vector regression (SVR) were assessed as candidate nonlinear models. Candidate patients were chosen for NGS sequencing based on the proportional difference between their true and predicted values from the original systemic exposure model. Variant prioritization used the Sequence Kernel Association Test.

Results: Atorvastatin linear model fit was statistically significantly improved by incorporating the selected concomitant medications, but rosuvastatin model fit was not. Predictive performance was not improved using GAMs or SVR compared to linear regression, likely due to small sample size. Three candidate genes and corresponding observed variants were identified and discussed as potential predictors of systemic rosuvastatin exposure.

Conclusion: Linear modelling of systemic atorvastatin exposure can be improved by incorporating concomitant medications. The feasibility of using nonlinear predictive models is limited by small sample size. Future research on newly identified interacting medication and genetic variants may provide new insights regarding underlying molecular mechanisms affecting systemic statin exposure.

Included in

Epidemiology Commons

Share

COinS