J.A. Chris Delaney
Warfarin is an anticoagulant medication that is used for the prevention and treatment of venous and arterial thrombotic complications. The evaluation of the risks associated with warfarin therapy, and its interaction with other drugs, poses important methodological challenges. We examined warfarin risks using data from the United Kingdom~Rs General Practice Research Database (GPRD).
First, we conducted a case-control study to examine the risk of gastrointestinal bleeding associated with warfarin use. We identified 4028 cases and 40171 matched controls from 2000 through 2005. Using conditional logistic regression, we found an increased risk of bleeding associated with warfarin use [adjusted odds ratio (OR) 2.15; 95% confidence interval (CI):1.81 to 2.54]. We also observed an increased risk due to drug-drug interactions between warfarin and other anti-thrombotic drugs. We also observed evidence of channelling bias as warfarin users were less likely to be prescribed other anti-thrombotic drugs.
Second, to attempt to deal with channelling bias, we re-analyzed this cohort using a Marginal Structural Model. This analysis produced a different estimate for the risk of bleeding associated with warfarin [OR 17.2; 95% CI: 6.5 to 37.7] than analysis with conditional logistic regression. The impact of effect modification on these estimates was then assessed with a Monte Carlo simulation study to determine if this was a plausible estimate.
Third, we created a cohort of patients given their first prescription of warfarin, ibuprofen, statins, or rofecoxib/celecoxib from 2001 through 2003 to study whether GPRD blood pressure data was of sufficient quality to model longitudinal increases in blood pressure as an adverse event. We compared different approaches to handling missing data including a mixed model and multiple imputation. A hypothesized increase in systolic blood pressure when initiating warfarin therapy was not supported with an observed reduction of 0.23 mmHg (95% CI:-0.78 to 0.31).
We concluded that it was possible to assess the increased risk of adverse events among warfarin users using a clinical database so long as care was taken in dealing with the limitations of these data.
