Abstract
Existing warfarin dosing methods do not accurately predict warfarin maintenance doses for patients in the lower or upper quartile of dose requirements. It was argued that this is related to the use of the international normalised ratio (INR) as a sole marker of anticoagulation for warfarin dose individualisation. The overarching premise of this thesis was that the coagulation proteins are on the causal path from warfarin dose to INR response and that a measure of coagulation protein response in addition to the INR will be helpful in the prediction of future anticoagulant response. The aim of this thesis was to apply pharmacometric methods to understand the coagulation kinetics underpinning the warfarin dose response and to introduce a new perspective to the prediction of anticoagulant response to warfarin.
A joint model was developed to quantify the influence of warfarin on all six vitamin K-dependent coagulation proteins (factors II, VII, IX, X, and proteins C and S) simultaneously. The full correlation structures that exist between parameters at the individual level and between residual errors of different coagulation proteins were accounted for. Of all the coagulation proteins considered, factor VII was found to have the shortest degradation half-life and will therefore be the first to reach a new steady-state following a perturbation introduced by warfarin.
Subsequently, the influence of coagulation proteins and their interactions on the INR was explored based on simulations from a mechanistic coagulation network model. A sensitivity analysis revealed that INR is most sensitive to factor VII and an isobologram analysis demonstrated that the presence of more than one coagulation protein deficiencies is redundant for INR effect. It was proposed that factor VII is the most influential on the INR and that the use of factor VII as a marker of anticoagulation (in addition to the INR) may improve the prediction of the anticoagulant response.
A factor VII-based method for the prediction of anticoagulant response to warfarin was developed based on a heuristic model-order reduction of the coagulation network model. The prediction method was shown to be associated with minimal bias and its use was illustrated using data from one typical simulated patient and two real patients supporting a proof-of-principle.
Finally, a framework for systematic evaluation of model assumptions was developed. In particular, a flowchart was proposed to evaluate assumptions based on the impact and the probability of assumption violation. The assumptions underpinning the pharmacometric analyses presented in this thesis were evaluated and used to illustrate the utility of the proposed framework.
In this thesis, both the top-down and bottom-up pharmacometric analyses were applied to explore the coagulation kinetics underpinning warfarin dose response. Standard methods such as population analysis, model simulations, isobologram analysis, and sensitivity analysis were employed. A heuristic model-order reduction method was experimented and seemed to work well although generalisation of the method to other settings requires prospective testing.
The work conducted in this thesis offered a new perspective on the prediction of anticoagulant response. The next step would be to extend the current method to the prediction of warfarin maintenance dose. This would require setting up and evaluating a dose individualisation algorithm (perhaps Bayesian) that incorporates a factor VII-INR bivariate response variable. Last but not least, a framework for systematic evaluation of assumptions was proposed. An important future step would be to apply the framework to a series of other settings to fully explore the utility and robustness of the framework to different model-building processes and model use settings.