Background: How the adoption of prediction models to decide which patient with atrial fibrillation (AF) to anticoagulate can affect prescription rates and outcomes is unclear.
Methods: We retrospectively analyzed data from Danish registries on patients with a first-time recorded AF from 2005 to 2010. We simulated the adoption of a decisional model based on the individual absolute risk reduction of stroke and absolute risk increase of bleeding with warfarin, as expected from the patient CHA2DS2-VASc and HAS-BLED, adjusted for a 0.6 relative value for bleeding versus stroke. We studied 3 different model versions and calculated for each of them the net benefit associated with its adoption, measured as the value-adjusted reduction in stroke and bleeding events at 1 year, compared with i) the actual practice, or ii) recommending warfarin consistently with the European Society of Cardiology (ESC) guidelines, irrespective of HAS-BLED.
Results: We included 41,455 patients; 31.9% actually received warfarin. The expected treatment rate with the model ranged from 21% to 87% according to the version used. The model version resulting into the highest treatment rate (i.e. treating any patient with CHA2DS2-VASc ≥ 1) was associated with the greatest net benefit (0.98; 95% credible interval 0.72-1.23), compared with the actual practice, with a 1/3 reduction in overall mortality, as with the adoption of ESC guidelines.
Conclusions: Preliminarily to a randomized impact study, our analysis suggests that individualizing anticoagulation for AF using a decisional model might have a clinical advantage over actual practice, and no added advantage over following ESC guidelines.
Keywords: Anticoagulation; Atrial fibrillation; Decision model; Personalized medicine.
Copyright © 2015 Elsevier Ireland Ltd. All rights reserved.