Key Points
The influence of known genetic variants on warfarin dose differs by race.
Race-specific pharmacogenetic algorithms, rather than race-adjusted algorithms, should be used to guide warfarin dosing.
Abstract
Warfarin dosing algorithms adjust for race, assigning a fixed effect size to each predictor, thereby attenuating the differential effect by race. Attenuation likely occurs in both race groups but may be more pronounced in the less-represented race group. Therefore, we evaluated whether the effect of clinical (age, body surface area [BSA], chronic kidney disease [CKD], and amiodarone use) and genetic factors (CYP2C9*2, *3, *5, *6, *11, rs12777823, VKORC1, and CYP4F2) on warfarin dose differs by race using regression analyses among 1357 patients enrolled in a prospective cohort study and compared predictive ability of race-combined vs race-stratified models. Differential effect of predictors by race was assessed using predictor-race interactions in race-combined analyses. Warfarin dose was influenced by age, BSA, CKD, amiodarone use, and CYP2C9*3 and VKORC1 variants in both races, by CYP2C9*2 and CYP4F2 variants in European Americans, and by rs12777823 in African Americans. CYP2C9*2 was associated with a lower dose only among European Americans (20.6% vs 3.0%, P < .001) and rs12777823 only among African Americans (12.3% vs 2.3%, P = .006). Although VKORC1 was associated with dose decrease in both races, the proportional decrease was higher among European Americans (28.9% vs 19.9%, P = .003) compared with African Americans. Race-stratified analysis improved dose prediction in both race groups compared with race-combined analysis. We demonstrate that the effect of predictors on warfarin dose differs by race, which may explain divergent findings reported by recent warfarin pharmacogenetic trials. We recommend that warfarin dosing algorithms should be stratified by race rather than adjusted for race.
Introduction
The recognition of racial differences in disease diagnosis, treatment, and outcomes has led the National Institutes of Health to encourage broad racial representation in developing the evidence base for treatment. For warfarin, the most widely used oral anticoagulant, investigators have met this charge through individual1-13 and collaborative efforts14-16 to identify factors that influence dose requirements. This body of work provides a robust case study for understanding race-related differences in drug response.
The significant interpatient variability in the dose requirements necessary for optimal anticoagulation poses a unique challenge for warfarin therapy.17 Investigations have identified clinical (eg, age and comedications) and genetic factors that explain a significant proportion of dose variability. Although several genes influence warfarin dose, single-nucleotide polymorphisms (SNPs) in cytochrome P450 2C9 (CYP2C9, the principal enzyme that metabolizes warfarin) and vitamin K epoxide reductase complex 1 (VKORC1, the target protein inhibited by warfarin) account for significant dose variability.18,19
This evidence underpins the prevailing hypothesis that genotype-guided therapy should improve dosing accuracy and this, in turn, should improve anticoagulation control measured as percent time in therapeutic range (PTTR). Two clinical trials, the European Pharmacogenetics of Anticoagulant Therapy (EU-PACT)20 and the Clarification of Oral Anticoagulation through Genetics (COAG),21 evaluated whether genotype-guided dosing improved PTTR, reporting incongruous results. Moreover, African Americans receiving genotype-guided dosing had lower PTTR than those receiving clinical-guided dosing.21
We hypothesize that the effect of clinical and genetic factors on warfarin dose differs by race and that compared with race-combined dosing models, race-stratified models improve prediction accuracy for both race groups. We use this as a framework to reconcile the evidence from observational and interventional warfarin pharmacogenetic studies and discuss implications of combining race groups in analysis.
Methods
Study population
Participants (≥20 years old) initiating warfarin with the target international normalized ratio of 2 to 3 were enrolled at the beginning of treatment in an inception cohort under the approval of the institutional review boards of the University of Alabama at Birmingham and Emory University.
Clinical and genetic variables
Patient demographics, indications for therapy, comorbidities, and medications were collected as previously reported.4-6,22 In addition to VKORC1 (rs9923231), CYP2C9 (*2 [rs1799853], and *3 [rs1057910]), we assessed CYP4F2 (rs2108622), African American–specific CYP2C9 SNPs (*5 [rs28371686], *6 [rs9332131], *8 [rs7900194], and *11 [rs28371685]), the CYP2C SNP (rs12777823), folylpolyglutamate synthase (FPGS; rs7856096), epoxide hydrolase 1 (EPHX1; rs1057140), γ-glutamyl carboxylase (GGCX; rs699664), and calumenin (CALU; rs2290228) polymorphisms.4,16,23,24 The assumption of Hardy-Weinberg equilibrium was met for all SNPs (P > .20).
Outcome variable
Dose (mg/day; log transformed to attain normality) was defined as the average maintenance dose after the attainment of 3 consecutive international normalized ratios in target range measured at least 2 weeks apart.
Statistical analyses
Analysis of variance was used to assess group differences for continuous variables and χ2 for categorical variables. To evaluate the effect of predictors on warfarin dose, we first restricted the multivariable linear regression analysis to include predictors (age, body surface area [BSA], smoking status, venous thromboembolism [VTE] vs non-VTE indication, amiodarone cotherapy, and CYP2C9*2, CYP2C9*3, and VKORC1) included in COAG (model 1). To test whether the effect of predictors differed by race, we first conducted multivariable analysis adjusting for race (with and without predictor-race interactions) and then conducted race-stratified analysis.
Next, we incorporated additional variables (not included in the COAG) that have been reported to influence warfarin dose (model 2). These include chronic kidney disease (CKD),22,25 cotherapy with statins and antiplatelets, CYP4F2, African American–specific CYP2C9 SNPs (*5, *6, *8, and *11), rs12777823, FPGS, EPHX1, GGCX, and CALU. We also included factors that showed significant racial differences in prevalence in our cohort and retained these factors in the model at a nominal P value of ≤ .2. All analyses were conducted with SAS version 9.3 at a bidirectional α level of .05.
Results
Table 1 presents characteristics of the 1357 participants (mean age, 61.0 ± 15.8 years; 43.9% African American, 56.1% European American). VTE and stroke were common indications for therapy among African Americans. Atrial fibrillation was the most common indication among European Americans. Compared with European Americans, African Americans were younger and more likely to be female and current smokers and have hypertension, diabetes, and severe CKD but less likely to have hyperlipidemia and use statins, antiplatelets, or amiodarone. The prevalence of CYP2C9(*2,*3), VKORC1, CYP4F2, EPHX1, and CALU variants was higher among European Americans. The prevalence of CYP2C9*8, rs12777823, and GGCX variants was higher among African Americans. CYP2C9 variants *5, *6, and *11 and the FPGS variant were only encountered in African Americans.
Table 1.
Clinical, genetic, and sociodemographic characteristics of participants by race
| Characteristics | European Americans* (n = 762) | African Americans (n = 595) | P |
|---|---|---|---|
| Age, y (mean ± SD) | 64.1 ± 15.2 | 57.0 ± 15.7 | <.001 |
| Height, inches (mean ± SD) | 67.9 ± 4.0 | 67.4 ± 6.6 | .08 |
| Weight, pounds (mean ± SD) | 192.8 ± 50.1 | 200.6 ± 51.1 | .005 |
| BSA, m2 (mean ± SD) | 2.0 ± 0.3 | 2.0 ± 0.3 | .17 |
| Female | 324 (42.5) | 336 (56.5) | <.001 |
| Current smoker | 72 (9.5) | 97 (16.3) | <.001 |
| Indication for warfarin therapy | |||
| Venous thromboembolism | 265 (34.8) | 314 (52.8) | <.001 |
| Stroke/transient ischemic attack | 33 (4.3) | 41 (6.9) | .04 |
| Atrial fibrillation | 403 (52.9) | 170 (28.6) | <.001 |
| Myocardial infarction | 12 (1.6) | 11 (1.9) | .69 |
| Peripheral arterial disease | 7 (0.9) | 7 (1.2) | .64 |
| Other | 42 (5.5) | 51 (8.6) | .03 |
| Comorbid conditions | |||
| Hypertension | 464 (61.6) | 425 (72.2) | <.001 |
| Hyperlipidemia | 410 (54.5) | 238 (40.4) | <.001 |
| Diabetes mellitus | 202 (26.8) | 224 (38.0) | <.001 |
| CKD† | |||
| eGFR ≥60 mL/min/1.73 m2 | 460 (60.5) | 378 (63.7) | |
| eGFR ≥30-59 mL/min/1.73 m2 | 256 (33.7) | 134 (22.6) | <.001 |
| eGFR <30 mL/min/1.73 m2 | 44 (5.8) | 81 (13.7) | |
| Concurrent medications | |||
| Statins‡ | 444 (58.3) | 291 (49.4) | .001 |
| Antiplatelets§ | 472 (61.9) | 322 (54.7) | .007 |
| Amiodarone | 99 (12.9) | 42 (7.1) | <.001 |
| Percentage of patients possessing ≥1 minor allele|| | |||
| CYP2C9*2 | 25.6 | 4.9 | <.001 |
| CYP2C9*3 | 12.6 | 2.3 | <.001 |
| CYP2C9 *5, *6, *11 | 0.0 | 3.4 | <.001 |
| CYP2C9 *8 | 1.1 | 13.9 | <.001 |
| VKORC1 | 60.2 | 18.5 | <.001 |
| CYP4F2 | 51.9 | 16.4 | <.001 |
| rs12777823 | 30.6 | 42.1 | <.001 |
| FPGS | 0.0 | 40.4 | — |
| EPHX1 | 52.7 | 29.3 | <.001 |
| GGCX | 55.6 | 86.9 | <.001 |
| CALU | 35.6 | 13.2 | <.001 |
Data are presented as n (%) of participants unless otherwise indicated.
eGFR, estimated glomerular filtration rate; SD, standard deviation.
Asians (n = 4; 0.3%) and Hispanics (n = 5; 0.4%) were combined with the European Americans group as non–African Americans were combined into 1 race group in COAG.
All eGFRs are based on National Kidney Foundation staging using the Modification of Diet in Renal Disease study equation. Patients were categorized into 3 categories: GFR ≥60 (no CKD or mild CKD stage 1 and 2), GFR = 30 to 59 (moderate CKD; stage 3), and GFR <30 (severe CKD; stage 4 and 5).
Statins included any of the 3-hydroxy-3-methyl-glutaryl–coenzyme A reductase inhibitors.
Antiplatelet agents included aspirin, clopidogrel, and dipyridamole as monotherapy or dual therapy.
CYP2C9*2, CYP2C9*3, CYP4F2, and VKORC1 were included as additive: 0 if no variants, 1 if heterozygous, and 2 if homozygous for the variant allele. CYP2C9*5, *6, and *11 together and CYP2C9*8, rs12777823, FPGS (rs7856096), EPHX1 (rs1051740), GGCX (rs699664), and CALU (rs2290228) were categorized as 0 if no variants and 1 if heterozygous or homozygous for the variant allele. Genotyping was not complete for some patients at the time of analysis, and therefore, genotype information is not available in 86 patients for CYP2C9, 828 patients for CYP2C9*8 (rs7900194 “A” allele), 57 patients for VKORC1 (rs9923231 “T” allele), 117 patients for CYP4F2 (rs2108622; A allele), and 118 patients for rs12777823 (A allele). Analysis of the influence of FPGS (rs7856096; “G” allele), EPHX1 (rs1051740; G allele), GGCX (rs699664; T allele), and CALU (rs2290228; A allele) was restricted to 290 patients.
We evaluated the impact of predictors included in the algorithm implemented in COAG (model 1). Dose changes associated with each predictor in race-combined and race-stratified analysis are presented in Table 2. To assess if the impact of predictors differed by race, we evaluated predictor-race interactions in the entire cohort. The influence of age, BSA, smoking, indication for therapy, amiodarone cotherapy, and CYP2C9*3 carriage did not differ by race. Per variant allele, possession of CYP2C9*3 variants resulted in similar dose reductions among African Americans and European Americans, respectively (34.6% vs 34.4%; interaction P value = .98). Possession of variant VKORC1 was associated with a significant dose decrease in both race groups; however, the dose reduction per variant allele was higher among European Americans (28.3% vs 18.6%, interaction P value = .002) compared with African Americans. Possession of CYP2C9*2 was associated with lower doses among European Americans, but not African Americans (20.7% vs 4.0% per variant allele, interaction P value < .001).
Table 2.
Effect of clinical and genetic predictors implemented in the COAG trial on warfarin dose requirements by race
| Variable | Race combined* | Race stratified | P§ (race × predictor interaction) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| European Americans† | African Americans‡ | |||||||||
| β | % Dose change (95% CI) | P | β | % Dose change (95% CI) | P | β | % Dose change (95% CI) | P | ||
| Intercept | 1.4564 | — | — | 1.4749 | — | — | 1.3163 | — | — | — |
| African American | −0.0992 | −9.44 (−13.57 to −5.11) | <.001 | — | — | — | — | — | — | — |
| Age, y | −0.0068 | −0.68 (−0.81 to −0.54) | <.001 | −0.0064 | −0.64 (−0.82 to −0.46) | <.001 | −0.0070 | −0.7 (−0.9 to −0.49) | <.001 | .67 |
| BSA, per m2 | 0.4219 | 52.48 (41.3 to 64.55) | <.001 | 0.4143 | 51.34 (37.33 to 66.78) | <.001 | 0.4411 | 55.44 (37.81 to 75.33) | <.001 | .73 |
| Current smoker | −0.0022 | −0.22 (−6.12 to 6.06) | .94 | −0.0046 | −0.45 (−9.08 to 8.99) | .92 | −0.0078 | −0.78 (−8.64 to 7.77) | .85 | .96 |
| VTE | 0.0512 | 5.26 (0.85 to 9.86) | .02 | 0.0574 | 5.91 (0.04 to 12.11) | .05 | 0.0374 | 3.81 (−2.68 to 10.73) | .26 | .65 |
| Amiodarone | −0.2245 | −20.1 (−25.11 to −14.76) | <.001 | −0.1906 | −17.36 (−23.39 to −10.85) | <.001 | −0.3146 | −26.99 (−35.24 to −17.7) | <.001 | .08 |
| CYP2C9*2|| | −0.1929 | −17.54 (−21.54 to −13.34) | <.001 | −0.2313 | −20.65 (−24.67 to −16.42) | <.001 | 0.0389 | 3.96 (−9.79 to 19.81) | .59 | <.001 |
| CYP2C9*3|| | −0.4183 | −34.19 (−38.5 to −29.57) | <.001 | −0.4221 | −34.43 (−38.92 to −29.61) | <.001 | −0.4246 | −34.6 (−46.11 to −20.63) | <.001 | .98 |
| VKORC1|| | −0.3016 | −26.03 (−28.45 to −23.54) | <.001 | −0.3330 | −28.32 (−30.92 to −25.63) | <.001 | −0.2058 | −18.6 (−24.21 to −12.57) | <.001 | .002 |
β denotes parameter estimates. Bold font denotes that the parameter estimates are significantly different by race.
CI, confidence interval; VTE, venous thromboembolism as primary indication for warfarin therapy (vs non-VTE indications such as atrial fibrillation).
The referent is a European American with wild-type genotype for the CYP2C9*2, CYP2C9*3, and VKORC1, nonsmoker, without VTE, and not on amiodarone.
The referent is a European American with wild-type genotype for the CYP2C9*2, CYP2C9*3, and VKORC1, nonsmoker, without VTE, and not on amiodarone.
The referent is an African American with wild-type genotype for the CYP2C9*2, CYP2C9*3, and VKORC1, nonsmoker, without VTE, and not on amiodarone.
P value for the difference in parameter estimates obtained by including interaction terms of predictors with race (African American vs European American) in the race-adjusted model.
CYP2C9*2, CYP2C9*3, and VKORC1 were included as additive: 0 if no variants, 1 if heterozygous, and 2 if homozygous for the variant allele.
Next, we incorporated additional variables (not included in the COAG; model 2) that have been reported to influence warfarin dose. Clinical factors evaluated were gender (P = .29), cotherapy with statins (P = .42), antiplatelet therapy (P = .47), and CKD (P < .001). As these predictors, with the exception for CKD, were not statistically significant in European Americans or African Americans, they were excluded from further analysis, as were smoking status (P = .92) and VTE (P = .23). Additional genetic variables included CYP2C9*8 (P = .69), FPGS (P = .15), EPHX1 (P = .49), GGCX (P = .21), and CALU (P = .74), which were not statistically significant among both European Americans and African Americans and were, therefore, excluded from further analysis. CYP2C9*3 and VKORC1 were significantly associated with dose change among both European Americans and African Americans. CYP2C9*2 (P < .001) and CYP4F2 (P = .004) were found to be significant only among European Americans, whereas rs12777823 (P < .001) was statistically significant only among African Americans. The results of the final model are shown in Table 3.
Table 3.
Effect of clinical and genetic factors on warfarin dose requirements by race with incorporation of additional predictors
| Variable | Race combined* | Race stratified | P§ (race × predictor interaction) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| European Americans† | African Americans‡ | |||||||||
| β | % Dose change (95% CI) | P | β | % Dose change (95% CI) | P | β | % Dose change (95% CI) | P | ||
| Intercept | 1.5456 | — | — | 1.5413 | — | — | 1.4355 | — | — | — |
| African Americans | −0.0686 | −6.63 (−11.08 to −1.96) | .006 | — | — | — | — | — | — | — |
| Age, y | −0.0069 | −0.68 (−0.81 to −0.56) | <.001 | −0.0062 | −0.62 (−0.79 to −0.45) | <.001 | −0.0071 | −0.71 (−8.69 to −5) | <.001 | .53 |
| BSA, per m2 | 0.4127 | 51.09 (40.10 to 62.94) | <.001 | 0.4006 | 49.27 (35.69 to 64.21) | <.001 | 0.4375 | 54.89 (8.26 to 14.95) | <.001 | .65 |
| CKD|| | −0.0860 | −8.24 (−10.94 to −5.47) | <.001 | −0.1156 | −10.92 (−14.54 to −7.15) | <.001 | −0.0668 | −6.46 (−10.37 to −2.38) | .002 | .10 |
| Amiodarone use | −0.2230 | −19.99 (−24.85 to −14.81) | <.001 | −0.1926 | −17.52 (−23.37 to −11.23) | <.001 | −0.2977 | −25.75 (−33.81 to −16.7) | <.001 | .13 |
| CYP2C9*2¶ | −0.2042 | −18.47 (−22.40 to −14.33) | <.001 | −0.2312 | −20.64 (−24.66 to −16.41) | <.001 | 0.0294 | 2.98 (−10.52 to 18.53) | .68 | <.001 |
| CYP2C9*3¶ | −0.4282 | −34.83 (−39.08 to −30.29) | <.001 | −0.4162 | −34.05 (−38.52 to −29.24) | <.001 | −0.4753 | −37.83 (−48.92 to −24.33) | <.001 | .52 |
| VKORC1¶ | −0.3098 | −26.64 (−29.02 to −24.18) | <.001 | −0.3416 | −28.94 (−31.49 to −26.29) | <.001 | −0.2231 | −19.99 (−25.53 to −14.05) | <.001 | .003 |
| CYP4F2¶ | 0.0448 | 4.58 (0.97 to 8.31) | .01 | 0.0573 | 5.89 (1.9 to 10.05) | .004 | 0.0122 | 1.23 (−6.32 to 9.38) | .76 | .26 |
| rs12777823# | −0.0789 | −7.59 (−11.32 to −3.70) | .0002 | −0.0229 | −2.26 (−7.44 to 3.21) | .41 | −0.1307 | −12.26 (−17.55 to −6.62) | <.001 | .006 |
| CYP2C9*5,6,11# | −0.1796 | −16.44 (−28.47 to −2.38) | .02 | — | — | — | −0.1304 | −12.23 (−25.08 to 2.83) | .11 | — |
β denotes parameter estimates. Bold font denotes that the parameter estimates are significantly different by race.
CI, confidence intervals.
The referent is a European American with wild-type genotype for the CYP2C9*2; CYP2C9*3; CYP2C9*5, *6, and *11; VKORC1; CYP4F2; and rs12777823 variants, without CKD, and not on amiodarone.
The referent is a European American with wild-type genotype for the CYP2C9*2, CYP2C9*3, VKORC1, CYP4F2, and rs12777823 variants, without CKD, and not on amiodarone.
The referent is a African American with wild-type genotype for the CYP2C9*2, CYP2C9*3, CYP2C9*5, *6 and *11,, VKORC1, CYP4F2 and rs12777823 variants, no CKD, and not on amiodarone.
P value for the difference in parameter estimates obtained by including interaction terms of predictors with race (African American vs European American) in the race-adjusted model.
CKD was based on National Kidney Foundation staging using the Modification of Diet in Renal Disease Study equation.2 Patients were categorized into 3 categories: eGFR ≥60 (no CKD or mild CKD stage 1 and 2), eGFR = 30-59 (moderate CKD; stage 3), and eGFR <30 (severe CKD; stage 4 and 5).
CYP2C9*2, CYP2C9*3, CYP4F2, and VKORC1 were included as additive: 0 if no variants, 1 if heterozygous, and 2 if homozygous for the variant allele.
CYP2C9*5, *6, and *11 together and rs12777823 were categorized as 0 if no variants and 1 if heterozygous or homozygous for the variant allele.
The influence of age, BSA, CKD, amiodarone cotherapy, and CYP2C9*3 carriage did not differ by race. Per variant allele, possession of CYP2C9*3 resulted in similar dose reduction among African Americans and European Americans (interaction P value = .52; Table 3). Possession of variant VKORC1 was associated with a significant dose decrease in both race groups; however, the dose reduction per variant allele was higher among European Americans (28.9% vs 19.9%, interaction P value = .003) compared with African Americans. Possession of CYP2C9*2 was associated with lower doses among European Americans, but not African Americans (20.6% vs 3.0% per variant allele, interaction P value < .001). Possession of variant rs12777823 was associated significantly lower dose among African Americans, but not in European Americans (12.3% vs 2.3%, interaction P value = .006). Possession of CYP4F2 variant resulted in 5.9% higher dose (per allele) in European Americans, but not in African Americans; however, the parameter estimates were not significantly different.
Table 4 presents the proportion of dose variability uniquely explained by clinical and genetic factors in race-combined and race-stratified analysis using dosing algorithms shown in Tables 2 and 3. The predictors included in the COAG algorithm explained a significant proportion of dose variability (R2 = 45.7%) in race-combined analysis, with genetic factors accounting for greater portion of the dose variability compared with the clinical factors (22.1% vs 16.1%, P < .001). In race-stratified analysis, these predictors accounted for a larger variability in European Americans compared with African Americans (R2 = 51.4% vs 29.3%, P < .001). Clinical factors accounted for a larger, but not statistically significant, dose variability in African Americans than European Americans (R2 = 21.5% vs 14.7%, P = .09). Genetic factors explained greater dose variability in European Americans compared with African Americans (R2 = 34.1% vs 7.0%, P < .001).
Table 4.
Proportion of dose variability uniquely explained by clinical and genetic in race combined and race-stratified analysis using dosing algorithms shown in Tables 2 and 3
| Model 1 | Model 2 | P | |
|---|---|---|---|
| Race combined | 45.7 | 48.3 | <.001 |
| Clinical | 16.1 | 17.4 | <.001 |
| Genetic | 22.1 | 23.5 | <.001 |
| Race stratified | |||
| European Americans | 51.4 | 54.0 | <.001 |
| Clinical | 14.7 | 16.4 | <.001 |
| Genetic | 34.1 | 34.6 | .009 |
| African Americans | 29.3 | 33.9 | <.001 |
| Clinical | 21.5 | 22.8 | .002 |
| Genetic | 7.0 | 10.0 | <.001 |
Model 1 included predictors implemented in the COAG study: age, BSA, smoking status, VTE (as primary indication for warfarin therapy [vs non-VTE indications such as atrial fibrillation]), amiodarone cotherapy, CYP2C9*2, CYP2C9*3, and VKORC1. BSA was calculated as [(weight in kg)0.425 × (height in cm)0.725]/139.2. CYP2C9*2, CYP2C9*3, and VKORC1 were included as additive: 0 if no variants, 1 if heterozygous, and 2 if homozygous for the variant allele. Model 2 included model 1 predictors and CKD, CYP4F2, African American–specific variants CYP2C9*5, *6, *11, and rs12777823.
Inclusion of additional predictors (CKD, CYP4F2, and African American–specific variants CYP2C9*5, *6, *11, and rs12777823) improved the proportion of dose variability in both race groups. In the race-combined analysis, clinical and genetic factors explained a significant proportion of dose variability (R2 = 48.3%). Genetic factors accounted for greater portion of the dose variability compared with the clinical factors (23.5% vs 17.4%, P < .001). In race-stratified analysis, these predictors accounted for a larger variability in European Americans compared with African Americans (R2 = 54.0% vs 33.9%, P < .001). Clinical factors accounted for a larger variability of dose in African Americans (R2 = 22.8% vs 16.4%, P = .13), whereas genetic factors explained greater dose variability in European Americans (R2 = 34.6% vs 10.0%, P < .001). The results of the final model are presented in Table 4.
Discussion
Warfarin, which remains the most widely used oral anticoagulant, has been the focus of significant pharmacogenetic investigation, including observational and interventional studies in racially diverse populations. Although observational studies have demonstrated a consistent and significant effect of clinical and genetic predictors on dose variability, interventional studies assessing whether genotype-guided therapy improves dosing accuracy and PTTR have demonstrated discordant results. We demonstrate that the effect of predictors on warfarin dose differs by race and recommend warfarin dosing algorithms should be stratified by race rather than adjusted for race.
Observational studies have demonstrated the effect of clinical and genetic predictors on warfarin dose among European Americans and African Americans and spurred the development of dosing algorithms. Among these, the warfarin dosing algorithm11 and the International Warfarin Pharmacogenetics Consortium (IWPC) algorithm14 are widely used. Both algorithms incorporate race as a predictor along with clinical and genetic factors. The ability of these algorithms to predict dose and consequently PTTR and bleeding risk has been evaluated. However, most of these studies had limited sample size,26-29 and all were conducted in cohorts wherein >95% of participants were of European ancestry.26-30 Most studies did not demonstrate improvement in PTTR with genotype-guided dosing.26-29 However, Anderson et al30 recently reported improved PTTR compared with standard dosing as part of routine care.
Two trials tested whether genotype-guided dosing improves PTTR. The EU-PACT trial20 randomized 455 participants (European ancestry, 98.6%; African ancestry, 0.9%; other ancestry, 0.5%) to receive either genotype-guided therapy (modified IWPC algorithm)14,31 or standard dosing. Compared with standard dosing, genotype-guided therapy improved PTTR (60.3% vs 67.4%, P < .001) at 12 weeks. The COAG trial21 randomized 1015 participants (67% European American, 27% African American, 6% Hispanic) to genotype-guided vs clinically guided warfarin dosing using the dose-initiation algorithm by Gage et al11 (days 1-3), followed by dose-revision algorithm (days 4-5).12 Genotype-guided dosing did not improve PTTR (45.4% vs 45.2%, P = .91) at 4 weeks and resulted in improved PTTR among European Americans (2.8%, P = .15) but lower PTTR in African Americans (−8.3%, P = .01).
These findings have fueled discussion regarding the utility of genotype-guided dosing, with investigators suggesting reasons for the null association with PTTR and the discordant findings by race in COAG. These include racial differences in CYP2C9*2, CYP2C9*3, and VKORC1 frequencies, lack of inclusion of African American–specific variation, differences in indication, and subtle stratification introduced by age across race. Moreover, the differences in PTTR improvement (7% vs 3%) among Europeans in EUPACT vs COAG have been attributed to differences in study design, initiation algorithms, and differences in comparator arms.32
We propose that differences in the proportional African representation can explain PTTR differences among COAG participants. As the algorithms adjust (rather than stratify) for race, the heterogeneity introduced by race sheds light on the divergent findings among African Americans vs European Americans in COAG.
Among COAG participants who achieved maintenance dose, genotype-guided dosing improved dose prediction during initiation (R2 = 48% vs 27%, P < .001) and revision (R2 = 69% vs 54%, P < .001), compared with clinically guided dosing. However, this was driven by improved dose prediction among European Americans receiving genotype-guided during dose initiation (R2 = 52% vs 21% for African Americans, P = .001) and dose revision (R2 = 75% vs 40% for African Americans, P < .001). In contrast, dose prediction was marginally better, although not statistically significant, among African Americans receiving clinically guided dosing during dose initiation (R2 = 33% vs 17% for European Americans, P = .08), but not during dose revision (R2 = 50% vs 51% for European Americans, P = .91). Moreover, genotype-guided dosing resulted in dispensation of doses that were significantly different (≥1 mg/day higher or lower) than required more often in African Americans during dose initiation (62% vs 42%, P = .002) and dose revision (50% vs 34%, P = .01) than in European Americans. In contrast, clinically guided dosing did not significantly influence the dispensation of doses that were different (≥1 mg/day higher or lower) than required in during dose initiation (61% European Americans vs 52% African Americans, P = .13) and dose revision (40% European Americans vs 37% African Americans, P = .71). In COAG, the genotype-guided algorithm improved dose prediction among European Americans but hindered dose prediction in African Americans.
The 43% African American composition of our cohort allowed robust assessment of the impact of clinical and genetic factors on warfarin dose by race. Concordant with previous reports, dose-prediction models explained greater dose variability among European Americans than African Americans. Clinical predictors account for a greater portion of variability in African Americans, whereas genetic predictors account for a greater portion of the variability in European Americans.
To reconcile the evidence from observational and interventional studies, it is important to remember that these algorithms were derived from cohorts composed largely of European Americans, with African Americans comprising a smaller proportion (IWPC, 9%; Gage et al, 15%).11,14 Therefore, the parameter estimates are potentially weighted toward race groups with higher representation. Second, the algorithms assume that predictors have the same impact for all race groups, assigning a fixed effect size (β coefficient) to each predictor regardless of race. Therefore, the differential effect of a predictor by race (if one exists) is attenuated. This attenuation is likely to occur in both race groups but may be more pronounced in the less-represented race group. In contrast, as 98.6% of EUPACT participants were of European descent, attenuation of parameter estimates (if any) was inconsequential.
Consider, for example, the effect of CYP2C9*2. The literature supports the CYP2C9*2- decreased dose association among European Americans.33,34 However, to our knowledge, the independent influence of CYP2C9*2 on warfarin dose in African Americans is not established. Most studies evaluating CYP2C9 variant-dose relationship in African Americans have assessed the combined effect of any variant (*2, *3, and others) vs none.1,3-11 Among African Americans, variant CYP2C9*2 was not associated with dose in a multicenter study (P = .84)16 or the Vanderbilt DNA repository (P = .51).34 Similarly, we show that CYP2C9*2 is associated with a lower dose in European Americans, but not African Americans. Implementing dose algorithms that adjust for race would result in the assignment of a fixed dose reduction in both races, ignoring the differential effect of CYP2C9*2 by race. In this context, for COAG participants with variant CYP2C9*2, dosing algorithms assigned a reduced dose for African Americans (when no decrease is needed). Among African Americans possessing CYP2C9*2 variants, our results indicate that the algorithm implemented in COAG would have assigned a lower-than-needed dose (by −0.9 mg/day; interquartile range [IQR], −2.21 to −0.12; P < .001).
Although possession of variant VKORC1 was associated with a significant dose decrease in both race groups, the dose reduction per variant allele was higher among European Americans compared with African Americans. Therefore assigning a fixed dose reduction for both race groups could result a larger-than-needed dose reduction for African Americans and smaller-than-needed dose reduction for European Americans. This could explain why, compared with clinical-guided dosing, genotype-guided therapy failed to improve dose prediction among African Americans in COAG. Moreover, the smaller-than-needed dose reduction for European Americans with CYP2C9*2 and VKORC1 variants may have attenuated the improved dose prediction with genotype-guided therapy in this race group.
The warfarin dosing algorithms used in COAG did not incorporate African American–specific CYP2C9 (*5, *6, *8, and *11) or rs12777823. This could explain the overdosing observed among African Americans receiving genotype-guided dosing in COAG. This has been highlighted by a recent study by Drozda et al,35 which showed that the algorithm used in COAG overestimates warfarin doses in African Americans possessing CYP2C9 (*5, *6, *8, and *11) variants. Our findings are consistent with this. The algorithm used in COAG by not incorporating CYP2C9 (*5, *6, *8, and *11) variants could have resulted in prescription of a higher-than-needed dose (+1.28 mg/day; IQR, 0.17 to 1.46; P = .005) among African Americans possessing these variants.
Although undiscovered at the time of COAG enrollment, inclusion of the variant rs12777823 would have resulted in dose reductions in ∼42% of African Americans.16 Located within the CYP2C gene cluster, rs12777823 is upstream of CYP2C18 on chromosome 10q23, likely affecting CYP2C9 activity, as indicated by its influence on the metabolism of the S-enantiomer of warfarin. The algorithm implemented in COAG would have assigned higher-than-needed dose (0.45 mg/day; IQR, −0.99 to 2.46; P = .02 for “AG” and 1.34 mg/day; IQR, 0.45 to 2.48; P < .001) for patients with rs12777823 AG and “GG” genotypes. This could explain the greater frequency of dispensation of doses that were significantly higher (≥1 mg/day) than required for African Americans. To what extent the inclusion of these variants and the improved dose prediction in African Americans could have influenced the findings of the COAG trial remains to be tested. Of note, possession of the variant rs12777823 has been associated with reduced dose requirements in African Americans (minor allele frequency = 25%), but not European Americans (minor allele frequency = 14%).16 As we show in our race-combined analysis, assigning a fixed effect would result in a 6% dose reduction for both African Americans (rather than the 12% reduction) and European Americans (in whom no adjustment is needed).
Improvement in dose prediction (for both race groups) can be achieved by using race-stratified algorithms. For example, in African Americans, the dosing algorithm predicted a greater portion of the dose variability compared with that predicted by Hernandez et al (34% vs 27%).36 Therefore, we recommend using race-stratified algorithms to guide dosing over race-adjusted algorithms, as they incorporate clinical and genetic factors influencing dose requirements in that race group and allow the use of race-specific effect size (parameter estimates) to personalize dosing.
Although we assessed differential effect of predictors among African Americans and European Americans, we recognize that these results may not be generalizable to Hispanics and Asians.37 Additional studies are needed to identify predictors that may differentially influence warfarin response in these race groups. Moreover, it is likely that incorporation of other genetic13,38,39 and clinical factors, not assessed in this study, may further improve dose prediction and alter the effect sizes of the predictors evaluated herein. We recognize this limitation.
Finally, these results bring into focus the broader implications related to racial heterogeneity in pharmacogenetic studies, demonstrating that the effect of predictors varies by race and that these effects are dependent on predictor prevalence and its biological effect (which cannot be presumed to be uniform). This highlights the need for adequate racial representation to facilitate identification of the differential impact of a predictor by race and enable adaptive algorithms that can incorporate race-specific effect size to personalize therapy.
Acknowledgments
The authors are grateful to all the patients that participated in the study. The authors thank our research nurses for their untiring efforts with patient recruitment and the medical faculty and staff of the Anticoagulation Clinic for their help with identification of potential participants. This study has contributed samples to the NINDS Human Genetics Resource Center DNA and Cell Line Repository (http://ccr.coriell.org/ninds). The NINDS repository sample numbers corresponding to the samples used are ND04466, ND04556, ND04604, ND04605, ND04626, ND04869, ND04907, ND04934, ND04951, ND05036, ND05108, ND05175, ND05176, ND05239, ND05605, ND05606, ND05701, ND05702, ND05735, ND06147, ND06207, ND06385, ND06424, ND06480, ND06706, ND06814, ND06871, ND06983, ND07057, ND07234, ND07304, ND07494, ND07602, ND07711, ND07712, ND08065, ND08596, ND08864, ND08932, ND09079, ND09172, ND09760, ND09761, and ND09809.
This work was supported in part by a grant from the National Institutes of Health National Heart Lung and Blood Institute (RO1HL092173), National Institute of General Medical Sciences (R01GM081488), and the National Institutes of Health Clinical and Translational Science Award (CTSA) program (UL1TR000165).
Footnotes
There is an Inside Blood Commentary on this article in this issue.
The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
Authorship
Contribution: N.A.L. and T. M. Beasley conceived and designed the study; N.A.L. acquired data; Q.Y., N.L., T. M. Beasley, A.S., and N.A.L. conducted statistical analysis; N.A.L., T. M. Beasley, T. M. Brown, and A.S. interpreted results; N.A.L., C.E.H., J.T., and A.S. drafted the manuscript; and N.A.L., T. M. Brown, and D.K.A. revised the manuscript for intellectual content.
Conflict-of-interest disclosure: The authors declare no competing financial interests.
Correspondence: Nita A. Limdi, Department of Neurology, University of Alabama at Birmingham, 1235 Jefferson Tower, 625 19th St South, Birmingham, AL 35294-0021; e-mail: nlimdi@uab.edu.
References
- 1.Cavallari LH, Langaee TY, Momary KM, et al. Genetic and clinical predictors of warfarin dose requirements in African Americans. Clin Pharmacol Ther. 2010;87(4):459–464. doi: 10.1038/clpt.2009.223. [DOI] [PubMed] [Google Scholar]
- 2.Cavallari LH, Perera M, Wadelius M, et al. Association of the GGCX (CAA)16/17 repeat polymorphism with higher warfarin dose requirements in African Americans. Pharmacogenet Genomics. 2012;22(2):152–158. doi: 10.1097/FPC.0b013e32834f288f. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Kimmel SE, Christie J, Kealey C, et al. Apolipoprotein E genotype and warfarin dosing among Caucasians and African Americans. Pharmacogenomics J. 2008;8(1):53–60. doi: 10.1038/sj.tpj.6500445. [DOI] [PubMed] [Google Scholar]
- 4.Limdi NA, Arnett DK, Goldstein JA, et al. Influence of CYP2C9 and VKORC1 on warfarin dose, anticoagulation attainment and maintenance among European-Americans and African-Americans. Pharmacogenomics. 2008;9(5):511–526. doi: 10.2217/14622416.9.5.511. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Limdi NA, Beasley TM, Crowley MR, et al. VKORC1 polymorphisms, haplotypes and haplotype groups on warfarin dose among African-Americans and European-Americans. Pharmacogenomics. 2008;9(10):1445–1458. doi: 10.2217/14622416.9.10.1445. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Limdi NA, McGwin G, Goldstein JA, et al. Influence of CYP2C9 and VKORC1 1173C/T genotype on the risk of hemorrhagic complications in African-American and European-American patients on warfarin. Clin Pharmacol Ther. 2008;83(2):312–321. doi: 10.1038/sj.clpt.6100290. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Perera MA, Gamazon E, Cavallari LH, et al. The missing association: sequencing-based discovery of novel SNPs in VKORC1 and CYP2C9 that affect warfarin dose in African Americans. Clin Pharmacol Ther. 2011;89(3):408–415. doi: 10.1038/clpt.2010.322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Schelleman H, Brensinger CM, Chen J, Finkelman BS, Rieder MJ, Kimmel SE. New genetic variant that might improve warfarin dose prediction in African Americans. Br J Clin Pharmacol. 2010;70(3):393–399. doi: 10.1111/j.1365-2125.2010.03709.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Schelleman H, Chen J, Chen Z, et al. Dosing algorithms to predict warfarin maintenance dose in Caucasians and African Americans. Clin Pharmacol Ther. 2008;84(3):332–339. doi: 10.1038/clpt.2008.101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Schelleman H, Chen Z, Kealey C, et al. Warfarin response and vitamin K epoxide reductase complex 1 in African Americans and Caucasians. Clin Pharmacol Ther. 2007;81(5):742–747. doi: 10.1038/sj.clpt.6100144. [DOI] [PubMed] [Google Scholar]
- 11.Gage BF, Eby C, Johnson JA, et al. Use of pharmacogenetic and clinical factors to predict the therapeutic dose of warfarin. Clin Pharmacol Ther. 2008;84(3):326–331. doi: 10.1038/clpt.2008.10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Lenzini P, Wadelius M, Kimmel S, et al. Integration of genetic, clinical, and INR data to refine warfarin dosing. Clin Pharmacol Ther. 2010;87(5):572–578. doi: 10.1038/clpt.2010.13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Cavallari LH, Vaynshteyn D, Freeman KM, et al. CYP2C9 promoter region single-nucleotide polymorphisms linked to the R150H polymorphism are functional suggesting their role in CYP2C9*8-mediated effects. Pharmacogenet Genomics. 2013;23(4):228–231. doi: 10.1097/FPC.0b013e32835e95c7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Klein TE, Altman RB, Eriksson N, et al. International Warfarin Pharmacogenetics Consortium. Estimation of the warfarin dose with clinical and pharmacogenetic data. N Engl J Med. 2009;360(8):753–764. doi: 10.1056/NEJMoa0809329. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Limdi NA, Wadelius M, Cavallari L, et al. International Warfarin Pharmacogenetics Consortium. Warfarin pharmacogenetics: a single VKORC1 polymorphism is predictive of dose across 3 racial groups. Blood. 2010;115(18):3827–3834. doi: 10.1182/blood-2009-12-255992. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Perera MA, Cavallari LH, Limdi NA, et al. Genetic variants associated with warfarin dose in African-American individuals: a genome-wide association study. Lancet. 2013;382(9894):790–796. doi: 10.1016/S0140-6736(13)60681-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Wittkowsky AK, Devine EB. Frequency and causes of overanticoagulation and underanticoagulation in patients treated with warfarin. Pharmacotherapy. 2004;24(10):1311–1316. doi: 10.1592/phco.24.14.1311.43144. [DOI] [PubMed] [Google Scholar]
- 18.Flockhart DA, O’Kane D, Williams MS, et al. ACMG Working Group on Pharmacogenetic Testing of CYP2C9, VKORC1 Alleles for Warfarin Use. Pharmacogenetic testing of CYP2C9 and VKORC1 alleles for warfarin. Genet Med. 2008;10(2):139–150. doi: 10.1097/GIM.0b013e318163c35f. [DOI] [PubMed] [Google Scholar]
- 19.McClain MR, Palomaki GE, Piper M, Haddow JE. A rapid-ACCE review of CYP2C9 and VKORC1 alleles testing to inform warfarin dosing in adults at elevated risk for thrombotic events to avoid serious bleeding. Genet Med. 2008;10(2):89–98. doi: 10.1097/GIM.0b013e31815bf924. [DOI] [PubMed] [Google Scholar]
- 20.Pirmohamed M, Burnside G, Eriksson N, et al. EU-PACT Group. A randomized trial of genotype-guided dosing of warfarin. N Engl J Med. 2013;369(24):2294–2303. doi: 10.1056/NEJMoa1311386. [DOI] [PubMed] [Google Scholar]
- 21.Kimmel SE, French B, Kasner SE, et al. COAG Investigators. A pharmacogenetic versus a clinical algorithm for warfarin dosing. N Engl J Med. 2013;369(24):2283–2293. doi: 10.1056/NEJMoa1310669. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Limdi NA, Beasley TM, Baird MF, et al. Kidney function influences warfarin responsiveness and hemorrhagic complications. J Am Soc Nephrol. 2009;20(4):912–921. doi: 10.1681/ASN.2008070802. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Daneshjou R, Gamazon ER, Burkley B, et al. Genetic variant in folate homeostasis is associated with lower warfarin dose in African Americans. Blood. 2014;124(14):2298–2305. doi: 10.1182/blood-2014-04-568436. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Suarez-Kurtz G, Botton MR. Pharmacogenomics of warfarin in populations of African descent. Br J Clin Pharmacol. 2013;75(2):334–346. doi: 10.1111/j.1365-2125.2012.04354.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Limdi NA, Limdi MA, Cavallari L, et al. Warfarin dosing in patients with impaired kidney function. Am J Kidney Dis. 2010;56(5):823–831. doi: 10.1053/j.ajkd.2010.05.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Burmester JK, Berg RL, Yale SH, et al. A randomized controlled trial of genotype-based Coumadin initiation. Genet Med. 2011;13(6):509–518. doi: 10.1097/GIM.0b013e31820ad77d. [DOI] [PubMed] [Google Scholar]
- 27.Jonas DE, Evans JP, McLeod HL, et al. Impact of genotype-guided dosing on anticoagulation visits for adults starting warfarin: a randomized controlled trial. Pharmacogenomics. 2013;14(13):1593–1603. doi: 10.2217/pgs.13.145. [DOI] [PubMed] [Google Scholar]
- 28.Anderson JL, Horne BD, Stevens SM, et al. Couma-Gen Investigators. Randomized trial of genotype-guided versus standard warfarin dosing in patients initiating oral anticoagulation. Circulation. 2007;116(22):2563–2570. doi: 10.1161/CIRCULATIONAHA.107.737312. [DOI] [PubMed] [Google Scholar]
- 29.Borgman MP, Pendleton RC, McMillin GA, et al. Prospective pilot trial of PerMIT versus standard anticoagulation service management of patients initiating oral anticoagulation. Thromb Haemost. 2012;108(3):561–569. doi: 10.1160/TH12-03-0159. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Anderson JL, Horne BD, Stevens SM, et al. A randomized and clinical effectiveness trial comparing two pharmacogenetic algorithms and standard care for individualizing warfarin dosing (CoumaGen-II). Circulation. 2012;125(16):1997–2005. doi: 10.1161/CIRCULATIONAHA.111.070920. [DOI] [PubMed] [Google Scholar]
- 31.Avery PJ, Jorgensen A, Hamberg AK, Wadelius M, Pirmohamed M, Kamali F EU-PACT Study Group. A proposal for an individualized pharmacogenetics-based warfarin initiation dose regimen for patients commencing anticoagulation therapy. Clin Pharmacol Ther. 2011;90(5):701–706. doi: 10.1038/clpt.2011.186. [DOI] [PubMed] [Google Scholar]
- 32.Kimmel SE, French B, Geller NL COAG Investigators. Genotype-guided dosing of vitamin K antagonists. N Engl J Med. 2014;370(18):1763–1764. doi: 10.1056/NEJMc1402521. [DOI] [PubMed] [Google Scholar]
- 33.Higashi MK, Veenstra DL, Kondo LM, et al. Association between CYP2C9 genetic variants and anticoagulation-related outcomes during warfarin therapy. JAMA. 2002;287(13):1690–1698. doi: 10.1001/jama.287.13.1690. [DOI] [PubMed] [Google Scholar]
- 34.Ramirez AH, Shi Y, Schildcrout JS, et al. Predicting warfarin dosage in European-Americans and African-Americans using DNA samples linked to an electronic health record. Pharmacogenomics. 2012;13(4):407–418. doi: 10.2217/pgs.11.164. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Drozda K, Wong S, Patel SR, et al. Poor warfarin dose prediction with pharmacogenetic algorithms that exclude genotypes important for African Americans. Pharmacogenet Genomics. 2015;25(2):73–81. doi: 10.1097/FPC.0000000000000108. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Hernandez W, Gamazon ER, Aquino-Michaels K, et al. Ethnicity-specific pharmacogenetics: the case of warfarin in African Americans. Pharmacogenomics J. 2014;14(3):223–228. doi: 10.1038/tpj.2013.34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Cavallari LH, Momary KM, Patel SR, Shapiro NL, Nutescu E, Viana MA. Pharmacogenomics of warfarin dose requirements in Hispanics. Blood Cells Mol Dis. 2011;46(2):147–150. doi: 10.1016/j.bcmd.2010.11.005. [DOI] [PubMed] [Google Scholar]
- 38.Voora D, Koboldt DC, King CR, et al. A polymorphism in the VKORC1 regulator calumenin predicts higher warfarin dose requirements in African Americans. Clin Pharmacol Ther. 2010;87(4):445–451. doi: 10.1038/clpt.2009.291. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Bress A, Patel SR, Perera MA, Campbell RT, Kittles RA, Cavallari LH. Effect of NQO1 and CYP4F2 genotypes on warfarin dose requirements in Hispanic-Americans and African-Americans. Pharmacogenomics. 2012;13(16):1925–1935. doi: 10.2217/pgs.12.164. [DOI] [PMC free article] [PubMed] [Google Scholar]