Abstract 169: The Role Of Clinical Prediction Factors On Anticoagulant Selection In Atrial Fibrillation
Background: Atrial fibrillation (AF) often benefits from the use of anticoagulants for prevention of stroke or systemic embolism. While novel oral anticoagulants have emerged as possible alternatives to warfarin, it is unknown how treatment selection is determined in practice with clinical guidelines still evolving. This study examined whether and to what extent anticoagulant selection has been driven by clinical predictions of stroke risk (treatment benefit) and bleeding risk (treatment harm) in real-world practice in the US. Methods: A nationwide database of commercial and Medicare Part D supplement claims from 2009-2011 was used to extract a cohort of non-valvular AF patients who were newly-initiating therapy after dabigatran availability in Oct 2010. Patients were excluded if they had claims associated with a reversible AF condition. Risk scores of ischemic stroke (CHADS2 and CHA2DS2-VASc) and bleeding (ATRIA) were used to examine associations with either warfarin or dabigatran use, calculated via claims in the outpatient pharmaceutical, inpatient medical, outpatient, and provider claims files. Baseline demographic and clinical characteristics were also measured as covariates, including concomitant diseases and medications. Multivariable log-binomial regression models assessed the association between each risk score and anticoagulant use, adjusting for the measured covariates. C-statistics were also used to examine the variation in treatment selection explained by inclusion of the risk scores. Results: In total, 37,401 patients were identified with 31% initiating dabigatran. New users of dabigatran were more likely to be younger, male, and have comorbidities. Patients at intermediate stroke risk (CHADS2 or CHA2DS2-VASc =1) were equally likely to receive warfarin and dabigatran (RR, 95% CI: 0.98, 0.93-1.02), while selection for warfarin was significantly associated with high ischemic stroke risk (CHADS2 or CHA2DS2-VASc ≥2) (RR, 95% CI: 0.87, 0.83-0.92). New users of dabigatran were significantly less likely to have high bleeding risk (ATRIA≥5) versus warfarin (RR, 95%: 0.70, 0.66-0.74). The c-statistic of the base model, which included the other measured covariates, was only marginally increased with the addition of any of the risk scores. Conclusions: Despite controlling for other patient characteristics, bleeding risk was strongly associated with the selection of a specific anticoagulant. However, the extent of selection explained by predictions of treatment harm was minimal. Providers appear to base anticoagulant selection on factors other than predictions of treatment benefit, which has implications for studying the anticoagulants’ comparative effectiveness.