Analysis of Mixed Models for Binary Longitudinal Data, with Application to Prescription Switching Patterns

Abstract

This thesis aims to study the statin use patterns of the Nova Scotia seniors population and the patients’ adherence to medication by applying a generalized linear mixed effect model (abbreviated as GLMM). Observations for a single subject will include the initial prescription and the sequence of transitions. The data can be modeled as short binary series, with tran- sition probabilities allowed to vary by subject. In this thesis, 10 sets of parameter values were run and the results were compared using tables and box plots. Mean Squared Error (MSE) and Estimated Bias (EB) are calculated to measure how close the estimated parameters are to the true values. For each parameter set, 10 and 100 simulations were run. We can make the conclusion that the generalized linear mixed effect model works well in the application of medication use patterns and the two separate GLMM models make sense.