Optimization of Anticoagulation Therapy by Secondary Analysis of Clinical Trial Data

Abstract

The objective of this thesis is conduct a secondary use of clinical data analysis using the ENGAGE clinical trial data set to calculate warfarin therapeutic dose for a large collection of clinical trial participants, and then test the accuracy of predictions of therapeutic dose using a collection of warfarin therapeutic dose prediction algorithms. We assessed demographic, physiological, clinical or genetic factors that may impact negatively on the predictions of the published warfarin therapeutic dose prediction algorithms for certain subpopulations of the general clinical trial sample population. Our results demonstrate that several of the prediction algorithms are very accurate for a large percentage of the population but that there exists distinct subpopulations, defined by specific factors such as genotype or risk scores, for which the generally accurate algorithms are dangerously inaccurate. Several previously unidentified clinical and drug-interaction covariates were identified that are important to the inaccurate predictions. Novel clinical variables including creatinine clearance, HAS-BLED, previous heart attack, and previous stroke—not include in standard algorithm and dosing protocols— were found to yield important differences in the computed and predicted therapeutic dose of warfarin. For genetic sensitive responders with HAS-BLED score greater than 3.0, over-prescription was seen in most cases with clinical or PGX dose algorithms, which might lead to an over-prescription for this subpopulation. Furthermore, we discover this subpopulation is more likely to experience any overt bleeding event within the first 90 days when compared to normal responders.