Warfarin-dosing algorithm based on a population pharmacokinetic/ pharmacodynamic model combined with Bayesian forecasting

Aims: To develop a novel warfarin-dosing algorithm based on a previous population pharmacokinetic/ pharmacodynamic (PK/PD) model with Bayesian forecasting to facilitate warfarin therapy. Materials & methods: Using information on CYP2C9 and VKORC1 genotypes, S-warfarin level, dose and international normalized ratio (INR) of prothrombin time, individual PK (apparent clearance of S-warfarin [CLs]) and PD (concentration resulting in 50% of E_max [EC₅₀]) parameters were determined by Bayesian forecasting for 45 Japanese patients. Maintenance doses were described by multiple linear regression using individually estimated PK/PD parameters and INR values. The validity of the model and a comparison with other dosing methods were evaluated by bootstrap resampling and a cross-validation method. Results: The plasma concentration of S-warfarin and INR were accurately predicted from individual PK/PD parameters. The following final regression model for maintenance dose was obtained; maintenance dose = 11.2 x CLs + 0.91 x EC ₅₀ + 2.36 x INR - 9.67, giving a strong correlation between actual and predicted maintenance doses (r² = 0.944). Bootstrap resampling and cross-validation showed robustness and a superior predictive performance compared with other dosing methods. On the other hand, the predictability without actual measurements (S-warfarin and INR values) and Bayesian inference was comparable to other dosing methods. Conclusion: A novel algorithm, based on the population PK/PD model combined with Bayesian forecasting, gave precise predictions of maintenance dose, leading to individualized warfarin therapy.