Development and Comparison of Warfarin Dosing Algorithms in Stroke Patients

Warfarin, the most commonly used oral anticoagulant in the world, is indicated for the prevention and treatment of thromboembolic events in patients with deep vein thrombosis, pulmonary embolism, atrial fibrillation and prosthetic heart valves. It is administered as a racemic mixture of S-warfarin and R-warfarin. S-warfarin is the more active isomer and has a greater therapeutic effect. These enantiomers are extensively metabolized by various cytochrome P450 enzymes. R-warfarin is mainly metabolized by cytochrome P450 1A2 (CYP1A2), CYP2C19, and CYP3A4, while S-warfarin is predominantly metabolized by CYP2C9.1 Both enantiomers affect the coagulation cascade by inhibiting the activity of vitamin K epoxide reductase complex 1 (VKORC1), thus interfering with the activation of clotting factors II, VII, IX, and X.2 However, despite its considerable benefit, warfarin is less frequently prescribed than it should be,3 because of it's relatively high adverse event rate and the difficulty in managing the therapy.4 Common adverse drug events (ADEs) arising from warfarin treatment include major and minor bleeding and hemorrhagic stroke.5 ADEs are also frequently due to insufficient therapy: ischemic stroke and venous thromboembolism.6 Warfarin, which has narrow therapeutic index, shows not only large inter-individual variations in dose response but also intra-individual variation. Because patients' clinical characteristics such as age, gender, body weight, concurrent medications, diet, co-morbidities and patient compliance level have shown to have large influence on warfarin dosing,7 frequent monitoring of its effect, as measured by the international normalized ratio (INR), is warranted.

CYP2C9 and VKORC1 have been identified as important genetic determinants of warfarin dosing. The most common CYP2C9 genotype among all ethnic is CYP2C9^*1 and found in about 80% of Caucasians8 and 93% of Korean.9 Lindh, et al.10 demonstrated that carriers of CYP2C9^*2 and CYP2C9^*3 alleles require less warfarin dose than carriers of wild type CYP2C9^*1 genotype. Difference in allelic frequencies are also observed with the most common single nucleotide polymorphisms (SNPs) in the VKORC1 gene, 1173C>T (rs9934438). Approximately, 35% of caucasians carry CC genotype, while only about 15% carry the TT genotype. However, most of the Korean showed TT genotype, but less than 1% carry CC genotype.11 Carriers of 1173CC and 1173CT genotype need 44% and 97%, respectively, more warfarin dose than carriers of 1173TT genotype.12 The clinical algorithms for determining warfarin dose containing clinical characteristics and pharmacogenetics information of the patients have been developed for more proper warfarin dose prediction. These algorithms are not intended to replace INR monitoring, but to increase the accuracy and reduce trial and error approach in warfarin dosing. According to International Warfarin Pharmacogenetics Consortium (IWPC) study,13 the algorithms using clinical and genetic information increase the accuracy in dose prediction than a fixed-dose approach or algorithms derived from clinical information. In addition, they are useful, especially in the patients who may be administered more or less than appropriate dose.13

Personalized dosing and INR monitoring are required, because response to warfarin is different depending on the indication and the state of the disease. The studies on warfarin dose assessment so far was mainly targeted at valvular heart disease, deep vein thrombosis and atrial fibrillation.13, 14, 15, 16, 17, 18, 19, 20, 21 On the other hand, warfarin is widely used to prevent the recurrence of stroke, which occupied the second place in the current causes of death in Korea, nevertheless, there are only a few studies to compare the predictive power of the dosing control based on a pharmacogenetics. This study reviewed prescribed dose and actual INR response in patients with stroke and compared the accuracy of 10 warfarin dose prediction algorithms based on the pharmacogenetics. In addition, warfarin dosing algorithm for Korean patients with stroke was developed to increase the quality of care for stroke patients.

Warfarin, the first human anticoagulant, remains the most commonly prescribed oral anticoagulant in the world. Warfarin exerts its effect by inhibiting the activity of VKORC1, thus interfering with the activation of vitamin K-dependent clotting factors II, VII, IX, and X.2 Warfarin is underutilized for stroke prevention. The Agency for Healthcare Policy and Research noted that physicians avoid to prescribe warfarin, because they are not familiar with techniques for administrating the drug safely and fear bleeding complication.3, 4 Warfarin therapy is challenging, because warfarin has narrow therapeutic index. In addition, it shows not only large inter-individual variations in dose response but also intra-individual variation. Because patients' clinical characteristics such as age, gender, body weight, concurrent medications, diet, co-morbidities and patient compliance level largely influence warfarin dosing,7 frequent monitoring of its effect, as measured by the INR, is warranted.

CYP2C9 and VKORC1 have been identified as important genetic determinants of warfarin dosing. Two prospective studies20, 27 on genotype-guided warfarin dosing predicted more accurately, resulting in reduction of dosing changes, minor bleeding complication, and time to reach therapeutic range. Although numerous warfarin dosing algorithms have been developed, their indications for warfarin usage were heterogeneous. Until now, there has been no consensus pharmacogenetic-guided algorithm.

In the present study, we developed an algorithm to provide a practical warfarin dosing for Korean patients with stroke. The warfarin dosing algorithm was developed on a homogeneous population and single disease indication for stroke, since warfarin have been underused for prevention of stroke and there are few studies on warfarin dosing algorithm focused on stroke patients. Because the distribution of warfarin dose is skewed, we created dosing algorithm for log transformation of doses, as evidenced by a mean percent deviation that was lower than that for both the raw doses square root of doses. Other variables including age and body weight showed normal distribution. We analyzed whether ten selected dosing algorithms, including the algorithm derived in the present study, could accurately predict warfarin dose in the study population, and found that the present algorithm demonstrated good correlation with actual dose, with coefficient of determination (R²) of 0.51. Algorithm in this study is consisted of five factors; age, body weight, and genotypes of CYP2C9 and VKORC1. While reduced incorporated factors are convenient for physicians to use, this algorithm performed better than Gage, et al.,14 Wu, et al.,21 and Huang, et al.19 The VKORC1 and CYP2C9 genotypes accounted for about 14% of the inter-individual variation of the maintenance daily warfarin dose. Because allele frequencies of VKORC1 and CYP2C9 are different depending on race, the R² values of these genes differ among studies.

Anderson, et al.,20 Ohno, et al.,16 and Sconce, et al.15 also showed good linear relationship with actual dose and predicted dose. However, the R² indicated the linearity of the association only, and the mean deviation from actual dose is a better measure of the algorithms' performance. Although the above three algorithms showed a good correlation with the actual dose in our study population, a better prediction of dosage was achieved by our model.

The algorithm devised by Cho, et al.18 was the latest warfarin dosing algorithm for Korean patients with atrial fibrillation and the best model for prediction of daily maintenance dose. This algorithm showed a good relationship between the actual dose and the predicted warfarin dose in our study population as reported previously. However, this algorithm was the worst performing algorithm by means of the mean deviation. The mean age, the strongest predictor of warfarin dose, was slightly older in cohort of Cho, et al., although its significance is unclear. Because these two algorithms developed for two different single disease indication; atrial fibrillation and stroke, these patients may be differently influenced by environmental factors such as coadministered drugs and comorbidities. Atrial fibrillation was indeed the most common indication of warfarin usage in this study. Likewise western countries,28 cervicocephalic artery dissections were common causes of stroke in young patients under 45. Cho, et al.18 reported that statin influences the daily dose of warfarin. Simvastatin, fluvastatin, and lovastatin potentiate warfarin's effect.24, 29 In our study, most patients were taking statin such as atorvastatin, pitavastatin, and rosuvastatin which do not affect the warfarin's effect. There is no correlation between daily dose of warfarin and statin status regardless of types of statin.

The algorithm derived in the present this study was less predictable among patients who required high doses of warfarin (≥7 mg/day). As this study was a small retrospective analysis with only a few patients requiring high doses, the results might have been skewed because of the individual patients variations. About 3% of the patients could have complication due to underdose. Of the three outliers, two patients were VKORC1 CT genotype. The other was TT type and he was taking antituberculosis drug. Rimfampin decreases INR increase via induction of hepatic metabolism of warfarin.24 Removal of these three data points did not improve the correlation coefficient for our algorithm (R²=0.51). However, the sample size was too small to make a conclusion on the efficacy of the dosing algorithms in this population.

In this study, we found 0 and 4% prevalence of CYP2C9^*2 and CYP2C9^*3, respectively, which is consistent with the report of Cho, et al. who also found no CYP2C9^*2 and a 8.5% prevalence of ^*3. For VKORC1, we found 82%, 17%, and 1% prevalence of VKORC1 TT, CT, CC genotype, respectively, which compares with the report of Cho, et al. who found 75.4%, 23.1%, and 1.5%, respectively.

Our study has several limitations. First, we did not have sufficient data to include potentially important factors such as vitamin K intake or compliance of administration even if we educated the patients when starting warfarin. However, the percentage of variability in warfarin dosing in our study is similar to that in other compared models, therefore, the effect of these variables is probably small. Second, because we investigated only one VKORC1 SNPs, requiring us to impute missing genotype for evaluating some models. Therefore, we substituted missing genotype based on linkage disequilibrium, which is generally reliable.25 Nevertheless, it may cause error that would lead to decrease of the accuracy of our model. Third, only 4% of the study population was younger than 40 years of age; therefore, additional models for stroke patients with younger age are needed, as age is important factor of prediction.

In order to further improve dosing algorithms, additional study is necessary in efforts to find new genes and SNPs contained with these genes that influence warfarin pharmacokinetics and pharmacodynamics. Although the incorporation of additional variables could improve predictive algorithm, the gains may be modest and probably do not justify the cost effectiveness and improvement of clinical outcome. In addition, studies on clinical utility of these pharmacogenetic-guided algorithms should be performed.

In conclusion, we developed warfarin dose prediction algorithm for patients with stroke, and it explained 51% of the variation in the daily maintenance warfarin dose. Further studies to elucidate clinical utility of genotype-guided dosing and find additional genetic association are necessary.