We also incorporated inpatient and outpatient diagnosis files to ascertain the history of cardiovascular disease, peripheral vascular disease, cerebrovascular signaling pathway disease, retinopathy, nephropathy, neuropathy, depression, chronic kidney disease, chronic liver disease, and chronic lung disease based on ICD-9-CM codes. Patients were classified as having chronic liver disease if they had at least one hospital admission or outpatient

visit with a diagnostic code of hepatitis B virus infection (ICD-9-CM codes 070.2x, 070.3x, V02.61), hepatitis C virus infection (070.41, 070.44, 070.51, 070.54, V02.62), chronic hepatitis (571.4), liver cirrhosis (571.2, 571.5, 571.6),

or alcoholic liver disease (571.0x, 571.1x, 571.2, 571.3x). A previous validation study using hospital administrative database reported a positive predictive value of 90% with this definition. 23 Covariate information included age, gender, and socioeconomic status (i.e., using monthly income as a proxy). Conditional logistic regression was used to estimate the crude and adjusted odds ratio (OR) and 95% confidence interval (CI) for the association between rosiglitazone/pioglitazone and cancer occurrence with “nonuse” as the reference group. Potential covariates, including socioeconomic status (monthly income level), Opaganib supplier diabetes complications and comorbidities at cancer diagnosis, other antidiabetic agents, antihypertensive medications, statin, and aspirin were examined. In the multivariate analysis, we adjusted for the use of short-acting human insulin, sulfonylurea, metformin, as these science antidiabetic agents were reported to be associated with cancer risks and could potentially confound the association. Other variables were chosen by using stepwise selection with P values < 0.10 for model entry and > 0.05 for removal. The association

between rosiglitazone/pioglitazone and individual cancer incidence was separately estimated after adjustment for potential confounders specific to that cancer type. In the dose- and duration-response analyses, we calculated the ORs for higher (≥120 DDD) and lower cumulative dose (<120 DDD) use, and for cumulative treatment duration ≥3, 2-3, 1-2, and ≤1 years. A two-sided P value < 0.05 was considered statistically significant. Approximately 15% participants claimed at lease one prescription for pioglitazone. Assuming a correlation coefficient for pioglitazone use between case and control was 0.5 and an ORs was 0.8, a study of 2,500 cases and 4 controls for each case would have a power ≥80% at α = 0.05.