| The primary goal of active pharmacovigilance is to detect the association between certain drugs and particular adverse drug reactions to these drugs through cohort data. Several statistical methods, specifically the logistic regression model, the logistic regression model with James-Stein shrinkage, the Cox model, and the random effects Cox model have been proposed to investigate drug-event association. In this thesis, for each method, we describe the underlying model, the estimation techniques, as well as their properties. We also apply these four models to a diabetes data set, which is extracted from a cohort database, in order to analyze the association between particular drugs of interest (Actos, Avandia, Metformin, Insulin, and Sulfonylurea) and certain adverse drug reactions (heart failure and acute myocardial infarction). We also consider the effects of age, gender, time since first exposure to a drug, and cumulative dose. |
