| dc.description.abstract | Patients with type 2 diabetes who initiate therapy with metformin often add a second-line agent to achieve more optimal glycemic control. Commonly used second-line medications such as sulfonylurea and insulin are highly efficacious in lowering blood glucose, but can lead to adverse effects such as hypoglycemia and weight gain. To further examine clinical outcomes of sulfonylurea and insulin, three studies were conducted using data from a national cohort of veterans with type 2 diabetes. First, to facilitate large-scale studies on hypoglycemia, I validated a composite definition for the identification of hypoglycemia events in a sample of cohort patients using chart reviews (n=321) as the reference standard. The positive predictive value (PPV) of the composite definition was 80% (80% for hospitalization events, 48% for emergency department events, and 96% for outpatient events). The emergency department definition was excluded in subsequent studies due to low PPV. Second, I examined if medication adherence, determined using pharmacy refill data, was a potential risk factor for hypoglycemia within a year of intensifying metformin treatment with a sulfonylurea. In this retrospective cohort study of 49,424 patients, the event rates of hypoglycemia were 20.9 (95% confidence interval [CI] 19.0, 23.0) and 22.4 (95% CI 20.6, 24.3) per 1000 person-years among low (<80%) and high (≥80%) adherence patients, respectively (adjusted hazard ratio [HR] 0.95, 95% CI 0.83, 1.09), showing no evidence that adherence was an important risk factor. Finally, I evaluated two mechanisms, change in BMI or hypoglycemia, through which insulin use may lead to an increased risk of death. Causal mediation analyses using Cox proportional hazards models were performed in a retrospective cohort of 28,892 patients. Change in BMI and hypoglycemia during the first 12 months of insulin or sulfonylurea use were each estimated to mediate 15% (95% CI -56, 258) and 1% (95% CI -69, 74) of the total effect of insulin on death (HR 1.14, 95% CI 0.83, 1.59) respectively, suggesting that the causal pathways do not explain a significant proportion of the effect. | |
