Quantifying the Impact of Drug Exposure Misclassification due to Restrictive Drug Coverage in Administrative Databases: A Simulation Cohort Study

Abstract

ObjectiveDrug exposure misclassification may occur in administrative databases when individuals obtain nonreimbursed drugs by paying “out-of-pocket” or via alternative drug coverage plans. We examined the apparent association between oral antidiabetic therapy and mortality by simulating the effects of restrictive drug coverage policies. MethodsPopulation-based cohort study of 12,272 new patients using oral antidiabetic agents were identified using the administrative databases of Saskatchewan Health, 1991 to 1996. We randomly misclassified 0% [base case], 10%, 25%, and 50% of known patients taking metformin according to either overt drug exposure (e.g., metformin users switched to nonusers) or time of metformin initiation (e.g., delayed capture of exposure); thereby simulating the use of a “non-formulary” or “special authorization” policy, respectively. We also simulated an age-dependent coverage policy, mimicking a policy restricted to seniors. ResultsMetformin use was associated with lower mortality compared with sulfonylurea use in the base case (adjusted hazard ratio [aHR] 0.88, 95% confidence interval [CI] 0.78–0.99) and the nonformulary simulations. The special authorization simulations demonstrated, however, an increasing relative mortality hazard of metformin versus sulfonylurea exposure: aHR 0.96, 95% CI 0.96–0.97 and aHR 1.34, 95% CI 1.31–1.37, for 10% and 50% delays in coverage capture respectively when 50% of metformin users were misclassified. Age-dependent drug coverage had a variable impact on mortality risk compared with the base-case cohort; however, a new-user simulation with a 1-year washout revealed consistent results to the base-case analysis. ConclusionRestrictive drug coverage policies may result in substantial drug exposure misclassification, potentially severely biasing the results of drug-outcome relationships using administrative databases.