Abstract 16253: Which is the Optimal Pharmacogenetic Algorithm for Guiding Warfarin Initiation? Comparison among 3 Leading Algorithms
Background: Warfarin therapy is limited by highly variable individual dose requirements, a narrow therapeutic window, and a high risk of adverse events associated with out-of-range prothrombin times. Several pharmacogenetic (PG) algorithms have been proposed to improve warfarin efficiency and safety. This study compared 3 leading algorithms: Gage (Washington University), International Warfarin Pharmacogenetics Consortium (IWPC), and CoumaGen-II for ability to predict stable maintenance dose (MD) in a large prospective study.
Methods: Patients (pts) enrolled in the CoumaGen-II study (NCT00927862) were evaluated. Individual patient clinical and genetic (CYP2C9, VKORC1) information was entered into previously published and on-line Gage and IWPC algorithms and the prospectively defined CoumaGen-II algorithm and MDs computed. Absolute differences between computed MDs and actual MDs were compared. An empiric dose of 5mg/d served as a virtual control. Differences between algorithms were compared using Friedman’s non-parametric analysis of related samples. Equivalence testing was evaluated using a tolerance of ±0.2 mg/d.
Results: Of 504 pts randomized, 444 reached a stable MD; 16 of these were excluded for missing data. Of 428 study pts, age averaged 61±14 y; 53% were female; 96% were white. Absolute differences (mg/d) between calculated and actual MD were small and not significantly different (p=0.17) among the 3 algorithms: Gage: median= 0.84, IQR=0.38-1.57; IWPC: median=0.83, IQR= 0.38-1.59; CoumaGen-II: median=0.76, IQR=0.32-1.58 mg/d. Equivalence testing showed all between-method comparisons to be equivalent after correcting for multiple comparisons (all p≤0.0018). In contrast, an empirically assumed fixed initial dose of 5mg/d was associated with a 1.43, IQR 0.57-2.43 absolute dose error, markedly worse (p<0.001) than with any of the 3 algorithms.
Conclusions: PG-guided warfarin dosing by each of 3 leading algorithms provided statistically equivalent estimates of MD and substantially improved predictions over fixed empiric dosing. Therefore, selection among these algorithms may be based on convenience of access and local preference, with the caveat that results may differ in non-Caucasian populations.
- © 2012 by American Heart Association, Inc.