Prediction of Warfarin Maintenance Dose in Han Chinese Patients Using a Mechanistic Model Based on Genetic and Non-genetic Factors

Background And Objectives: Many attempts have been made to predict the warfarin maintenance dose in patients beginning warfarin therapy using a descriptive model based on multiple linear regression. Here we report the first attempt to develop a comprehensive mechanistic model integrating in vitro-in vivo extrapolation (IVIVE) with a pharmacokinetic-pharmacodynamic model to predict the warfarin maintenance dose in Han Chinese patients. The model incorporates demographic factors [sex, age, body weight (BW)] and the genetic polymorphisms of cytochrome P450 (CYP) 2C9 (CYP2C9) and vitamin K epoxide reductase complex subunit 1 (VKORC1).

Methods: Information on the various factors, mean warfarin daily dose and International Normalized Ratio (INR) was available for a cohort of 197 Han Chinese patients. Based on in vitro enzyme kinetic parameters for S-warfarin metabolism, demographic data for Han Chinese and some scaling factors, the S-warfarin clearance (CL) was predicted for patients in the cohort with different CYP2C9 genotypes using IVIVE. The plasma concentration of S-warfarin after a single oral dose was simulated using a one-compartment pharmacokinetic model with first-order absorption and a lag time and was combined with a mechanistic coagulation model to simulate the INR response. The warfarin maintenance dose was then predicted based on the demographic data and genotypes of CYP2C9 and VKORC1 for each patient and using the observed steady-state INR (INRss) as a target value. Finally, sensitivity analysis was carried out to determine which factor(s) affect the warfarin maintenance dose most strongly.

Results: The predictive performance of this mechanistic model is not inferior to that of our previous descriptive model. There were significant differences in the mean warfarin daily dose in patients with different CYP2C9 and VKORC1 genotypes. Using IVIVE, the predicted mean CL of S-warfarin for patients with CYP2C9*1/*3 (0.092 l/h, n = 11) was 57 % less than for those with wild-type *1/*1 (0.215 l/h, n = 186). In addition, *1/*1 patients needed about 1 week to reach steady state, whereas *1/*3 patients needed about 2 weeks. In terms of the predicted INRss values, only ten patients had INRss values outside the expected therapeutic range (1.5-2.8). To evaluate our mechanistic model, we predicted the warfarin maintenance dose for 183 patients and explained 42 % of its variation, which is comparable to our previous prediction using a descriptive model based on multiple linear regression. The mean predicted/observed warfarin doses (mg/day) for different combinations of CYP2C9 and VKORC1 genotypes were 1.54/3.75 (n = 1) for *1/*1 and GG, 3.33/3.66 (n = 36) for *1/*1 and AG, 2.31/2.41 (n = 136) for *1/*1 and AA, and 1.56/1.69 (n = 10) for *1/*3 and AA, respectively. Sensitivity analysis indicated BW and genetic polymorphisms of CYP2C9 and VKORC1 were important factors affecting the warfarin maintenance dose in the study population.

Conclusion: The mechanistic model reported is the first to integrate IVIVE with a pharmacokinetic-pharmacodynamic model to describe the association of the warfarin maintenance dose with sex, age, BW and the genotypes of CYP2C9 and VKORC1. The model was effective in predicting S-warfarin clearance and in simulating its plasma concentration-time curve in a cohort of Han Chinese patients. In addition, the model accurately predicted the INR response and warfarin maintenance dose in a cohort of Han Chinese patients.