A population pharmacokinetic model for s-warfarin: application of a mixture model to determine genotype/phenotype.
Matthews, Ivan and Aarons, Leon
School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Manchester, UK.
Objectives: The vitamin K antagonist warfarin is an important drug in the treatment of thromboembolic disorders. The response of individuals to warfarin has been shown to vary widely. The polymorphic enzyme CYP2C9 is largely responsible for S-warfarin metabolism. The allelic variants CYP2C9*2 and CYP2C9*3 are associated with reduced activity and increased risk of haemorrhage especially at the initiation of therapy. The aims of this study were to predict the extent to which CYP2C9 polymorphisms alter the pharmacokinetics (PK) of warfarin and whether NONMEM can be used to correctly predict the genotype/phenotype of an individual using a mixture model.
Methods: Simcyp® version 5 was used to generate a virtual population of 100 patients and their individual PK profiles. The resulting patient demographics were; age 20-80 years, sex 50% male, weight 49-114kg and poor metaboliser (PM) phenotype 5%. They received 10mg of S-warfarin orally once a day for 7 days. A one compartment first order absorption PK model was fitted to the data using the ADVAN2 TRANS2 subroutine from the NONMEM library. An allometric weight model was applied to standardise V and CL using a standard weight (WTSTD) of 70 kg. In order to detect the polymorphism a mixture model with two CL subpopulations corresponding to a phenotype of PMs and extensive metabolisers (EM) was tested. Where PMs are carriers of the CYP2C9*2 or *3 alleles. The results were then compared to a categorical model which uses the phenotypic data provided by Simcyp® as a covariate.
Results: The mixture model was not able to assign any of the PMs to the appropriate group. When constrained, the value of the fractional covariate effect of CYP2C9 (FCYP) on CL tended towards 1 (no effect) and also an under prediction of the mixing proportion (no PMs). When left unconstrained the value of FCYP exceeded 1 creating an unidentified (by the CYP2C9 covariate data) group of ultra high metabolisers. Despite this the fit and population parameter estimates of the mixture model were similar to those of the categorical model except for the value of FCYP.
Conclusion: The mixture model was not efficient at assigning the appropriate phenotype to the patients.