Abstract 16531: Identifying Warfarin-Drug Interactions in Administrative Registries by Data Mining- A Pilot Study Focusing on Bleedings Requiring Hospitalizations
Background: Knowledge about drug-drug interactions mainly arise from preclinical trials, adverse drug reports, or is based on knowledge of pharmacological action. Methods enabling agnostic registry-based searches for drug-drug interactions are warranted to increase treatment safety. We investigated whether known warfarin-drug interactions could be identified using hospitalization-requiring bleedings as a prototype.
Methods: We applied random forest models to a nationwide administrative dataset of patients with nonvalvular atrial fibrillation treated with warfarin. First-time bleeding between 5 and 30 days after a novel prescription were defined as cases, and a control group of equal size was created using random samples of non-cases. The modeling was repeated twenty times (with resampled control groups) to evaluate the robustness. Drug groups were investigated on a 5 digits Anatomical Therapeutic Chemical (ATC) classification system level.
Results: We analyzed 779,764 novel prescriptions in 75,139 patients. 353 drug groups and 7,033 cases were investigated. The drug groups having the strongest associations with first time bleeding in warfarin treated patients are illustrated in the Figure.
Conclusion: We were able to identify known warfarin-drug interactions using bleeding as an endpoint. This suggests that data mining may be a sensitive method for investigating drug-drug interactions without a priori hypothesis.
Author Disclosures: P.W. Hansen: None. L. Clemmensen: None. T.S. Sehested: None. E.L. Fosbøl: None. C. Torp-Pedersen: Research Grant; Modest; Bristol-Myers Squibb. Consultant/Advisory Board; Modest; Cardiome, Merck, Sanofi, Daiichi. G.H. Gislason: Research Grant; Modest; Bristol Meyers Squibb, Boehringer-Ingelheim, Bayer, Pfizer, AstraZeneca. C. Andersson: None.
- © 2016 by American Heart Association, Inc.