Current Models for Predicting Warfarin Maintenance Dose Do Not Work in an Ambulatory Venous Thromboembolism Patient Population.

Warfarin is the most widely used anticoagulant for the treatment of conditions such as deep vein thrombosis and pulmonary embolism. Warfarin has a narrow therapeutic index, and individual patient response to the drug is highly variable. Models to predict the maintenance dose have been published but not validated in independent populations (Kamali et al, Sconce et al, and Gage et al). Most of the prior data was derived in patients with atrial fibrillation and these may not be applicable to our predominantly ambulatory VTE population. The prior models incorporate a variety of factors that have been predicted to have an effect on warfarin dosing, such as age, height, and polymorphisms in the vitamin K epoxide reductase complex subunit 1 (VKORC1) and cytochrome P450 (CYP450) 2C9 enzyme. We enrolled 51 consecutive VTE patients on stable doses of warfarin as defined in the prior studies. As in those studies we excluded patients with known or suspected non-compliance, known liver disease and known congestive heart failure. All previously suggested predictive variables were collected. Mean warfarin dose, mean INRs, and gene polymorphisms were determined by previously described standard methods. In addition we are testing the recently identified coding VKORC1 Asp36Tyr polymorphism which predisposes to warfarin resistance. This study aimed to examine the accuracy of three previous models to predict individual warfarin dosing in pursuit of the development of a more accurate regression model. Preliminary results with the first 51 patients, through correlation analysis, indicate that none of the previous models provide an accurate prediction of warfarin maintenance dose. The mean maintenance dose of the patient population was 6.23mg. The doses predicted by the previous models were significantly lower at 4.4mg, 5.3mg, and 4.5mg for Kamali, Gage, and Sconce, respectively. The Pearson correlation coefficients were 0.047, 0.27, and 0.50, respectively. A preliminary model has been developed with a Pearson correlation coefficient of r=0.78, and a p-value <0.0001. The R2 value for the model was 61.5%. We have now recruited over 290 patients and are analysing the data. We will report more precise data on correlations with the other models and if a new model with a higher R2 can be determined it will be reported.