Abstract
Oral anticoagulants are used to treat and prevent blood clots. All anticoagulants carry the risk of bleeding if the systemic exposure is too high, while inadequate exposure will increase the risk of thrombosis. Therefore, the safe and effective use of all oral anticoagulants will require dose individualisation and monitoring. The overarching goal of this thesis is to critically evaluate and explore dose individualisation methods for warfarin and dabigatran therapy to improve patient outcomes.
For warfarin, methods for predicting the maintenance dose were investigated. Specifically, Chapter 2 investigates the predictive performance of a Bayesian dose individualisation tool for warfarin. It was found that the maintenance dose was over-predicted especially in patients requiring higher daily doses and further studies into the source of bias were conducted. Chapter 3 further evaluates whether published warfarin maintenance dose prediction algorithms can accurately predict the observed maintenance dose in patients who require ≥7 mg daily (the upper quartile of dose requirements). A systematic review and meta-analysis was conducted to answer this question. It was found that all warfarin dosing algorithms included in the study under-predicted the maintenance dose in this group of patients.
One common metric to measure predictive performance of a model is the mean prediction error, which is a measure of bias. The work conducted in Chapter 2 and 3 suggests that the mean prediction error may not capture non-constant bias. This is when the predictions systematically deviate away from the line of identity in one direction in relation to the observed data. Chapter 4 proposes new method to assess predictive performance to analyse non-constant systematic deviation from the line of identity. The proposed method is not specific to warfarin, but can be applied to the analysis of predictive performance in general.
For dabigatran dosing, aspects of concentration monitoring as a means of determining a suitable dosing rate were explored. In Chapter 5, an assay using liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) was developed to measure all active entities of dabigatran concentrations in human plasma. The assay was used to measure dabigatran concentrations collected from a previous study. A de novo population pharmacokinetic model was not pursued in the first instance as the data were fairly sparse. Instead, the measured concentrations were used in Chapter 6 in a simulation based study to select an appropriate prior population pharmacokinetic model that might be used in a future Bayesian dose individualisation method for dabigatran. The overall intention of Chapter 6 was to develop a Bayesian dose individualisation method for dabigatran.
In conclusion, this thesis has identified the limitations of current methods for predicting warfarin maintenance dose and has explored dabigatran concentration monitoring as a means of improving dabigatran dosing. Models for predicting warfarin maintenance dose were critically evaluated and it was found that all existing models can not accurately predict the maintenance dose in patients requiring higher daily doses. An improvement in the method to assess predictive performance was proposed. The work conducted in this thesis on dabigatran dosing provides the basis for future research to individualise dosing and monitoring using population pharmacokinetic models.