Background: Warfarin is used to treat and prevent blood clots. It is one of the most difficult drugs to dose accurately, with daily maintenance doses varying by more than tenfold between patients. Evidence from pharmacogenetic studies indicates the importance of CYP2C9 and VKORC1 polymorphisms in predicting warfarin dose variability. However, this information is of limited value once INR measures are available. The INR is commonly used in clinical practice to individualise warfarin dosage. It acts as a biomarker for warfarin effect on coagulation and will, therefore, reflect all sources of variability on warfarin PKPD. We explore optimal INR sampling designs to maximise information about warfarin PKPD and to provide a basis for Bayesian dose individualisation. A previously published PKPD model for the effect of warfarin on INR by Hamberg et al 2007 [ref] provides the starting point for this research.
Aims and objectives: The primary aim of this research was to develop an optimal INR sampling design for incorporation into TCIWorks to provide the basis for Bayesian dose individualisation. This research had the following specific objectives;
- To externally evaluate a previously published PKPD model for warfarin by Hamberg et al 2007 [ref]
- To develop an optimal design to estimate the PKPD parameters for the Hamberg model
- To develop an optimal design to estimate the maximum a-posteriori (MAP) parameters for the Hamberg model.
- To adapt a model that could be used with TCIWorks with a recommended INR sampling design.
Methods: 1. A previously published warfarin PKPD dataset from O’Reilly et al [ref, ref] was used as an external evaluation of the Hamberg model. The Hamberg model was compared to our own model developed specifically for the O’Reilly data using VPCs. 2. An optimal design for the full PKPD model was developed using WinPOPT. This design was tested using simulation and estimation via NONMEM. 3. An optimal design for the MAP estimators in particular warfarin CL and the Emax and EC50 of warfarin on the VKORC1 enzyme were optimised. This design was assessed by simulation and estimation.
Results: 1. The PK model developed from Hamberg provided an acceptable description of the O’Reilly data. The PKPD model is being assessed. Items 2-4 are currently underway.
Conclusions: Dose individualisation using Bayesian methodology is a logical step to improve INR control. A model and design appropriate for clinical use have been developed as part of this project.
References
- Hamberg et al 2007. Clin Pharmacol Ther 81(4): 529-538.
- O’Reilly et al 1963. J Clin Inv 42(10): 1542-1551.
- O’Reilly & Aggeler 1969. Circulation 38: 169-177.