| The global diabetes epidemic is expected to increase from about 285 million people in 2010 to 439 million people by 2030. Due to the chronic nature of Type II diabetes, once patients begin treatment with oral hypoglycemic agents, they are usually treated for the remainder of their lives.;This dissertation examines the association between a newer class of oral hypoglycemic agent, thiazolidinidiones, and microvascular and macrovascular outcomes. These associations are particularly susceptible to confounding by indication bias because patients with severe diabetes are preferentially prescribed thiazolidinidiones, and are also independently at an increased risk of adverse outcomes. Using the United Kingdom-based General Practice Research Database, this dissertation employs several methods to address confounding by indication bias.;Study 1 develops and validates an algorithm to predict likely treatment with a thiazolidinidione during the drugs' pre-market period. Identifying likely treated patients could provide information about the baseline risk of adverse events, and consequently the likelihood and extent of confounding by indication bias expected in post-marketing surveillance studies. However, it was difficult to predict likely TZD treatment: our algorithm resulted in a sensitivity of 72%, specificity of 36%, positive predictive value of 20%, and negative predictive value of 85%.;Study 2 examines the association between thiazolidinidiones, compared to other oral hypoglycemic agents, and the risk of microvascular outcomes (retinopathy, neuropathy, and nephropathy). We observed no association with the risk of any microvascular disease (HR = 1.0; 95% CI = 0.97--1.0). However, when examining likely treatment compared to unlikely treatment in the pre-market period, we found an increased risk of any microvascular disease (HR = 1.5; 95% CI = 1.4--1.6), indicating that the null post-market association could actually reflect a protective effect, despite underlying confounding by indication.;Study 3 estimates the association between thiazolidinidiones and myocardial infarction using three study designs: nested case-control study, marginal structural model, and instrumental variable analysis. We found a small increased risk compared to treatment with any other oral hypoglycemic agent (RR = 1.4; 95% CI = 0.99--2.1); patients treated with thiazolidinidione monotherapy appeared to be at a greater risk than patients treated with thiazolidinidione polytherapy. |
