Original article 501

Prediction of stable acenocoumarol dose by a pharmacogenetic algorithm Enrique Jiménez-Varoa,c, Marisa Cañadas-Garrea, María J. Gutiérrez-Pimentelb and Miguel Á. Calleja-Hernándeza,c Aim To develop an acenocoumarol (ACN) dosing algorithm for patients with atrial fibrillation or venous thromboembolism, considering the influence on the stable ACN dose of clinical factors and gene polymorphisms, including CYP2C9*2/*3, VKORC1, CYP4F2*3, ABCB1, APOE, CYP2C19*2/*17, and GGCX. Methods and results A retrospective observational study was carried out to obtain clinical and pharmacogenetic dose algorithms by multiple linear regression of results in a cohort of 134 patients under treatment with a stable ACN dose for atrial fibrillation or venous thromboembolism and to test them in an independent validation cohort of 30 patients. The pharmacogenetic dosing algorithm included CYP2C9, VKORC1, and APOE, which explained 56.6% of the variability in the stable ACN dose. Lower deviation from the stable dose and increased accuracy were shown by the pharmacogenetic algorithm, which correctly classified 67% of patients with a deviation of up to 20%.

Conclusion The variability in the stable ACN dose was better explained by a pharmacogenetic algorithm including clinical and genetic factors (CYP2C9, VKORC1, and APOE) than by a clinical algorithm, providing a more accurate dosage prediction. Pharmacogenetics and Genomics 24:501–513 © 2014 Wolters Kluwer Health | Lippincott Williams & Wilkins. Pharmacogenetics and Genomics 2014, 24:501–513 Keywords: acenocoumarol, algorithm, APOE, CYP2C9, pharmacogenetics, VKORC1 a Pharmacogenetics Unit, UGC Provincial de Farmacia de Granada, Instituto de Investigación Biosanitaria de Granada, Hospitales Universitarios de Granada Avda. Fuerzas Armadas, 2, bHaematology Department, Complejo Hospitalario de Granada and cDepartment of Pharmacology, Faculty of Pharmacy, University of Granada, Campus Universitario de Cartuja, Granada, Spain

Correspondence to Marisa Cañadas-Garre, PhD, Pharmacogenetics Unit, UGC Provincial de Farmacia de Granada, Instituto de Investigación Biosanitaria de Granada, Hospitales Universitarios de Granada Avda. Fuerzas Armadas, 2, 18014 Granada, Spain Tel: + 34 958 020 108; fax: + 34 958 020 004; e-mail: [email protected] Received 19 March 2014 Accepted 8 July 2014

Introduction Vitamin K antagonists (VKAs), such as acenocoumarol (ACN) and warfarin, are the most widely prescribed oral anticoagulants to treat and prevent stroke and systemic embolism in atrial fibrillation (AF) and venous thromboembolism (VTE) patients. Variability in the initial dose needed to achieve stable anticoagulation depends on clinical (e.g. age, sex, weight, concomitant drugs), environmental (e.g. dietary vitamin K intake, smoking status, alcohol intake), and genetic factors. The most widely studied genetic variations correspond to genes encoding CYP2C9 (CYP2C9*2: Arg144Cys, rs1799853; CYP2C9*3: Ile359Leu, and rs1057910) and vitamin K epoxide reductase complex 1 (VKORC1) (1639G > A, rs9923231, and 1173C > T, rs9934438) [1–8]. It is well documented that clinical and genetic variables account for at least 50% of the interindividual variability in the required dose of VKAs [1–5]; in particular, VKORC1 (rs9923231) and CYP2C9 (rs1057910 and rs1799853) have been found to be responsible for ∼ 30% of this variability [6–8]. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's website (www.pharmacogeneticsandgenomics.com). 1744-6872 © 2014 Wolters Kluwer Health | Lippincott Williams & Wilkins

The initial dose adjustment of VKAs is mainly performed by experienced physicians on the basis of clinical parameters and modified by titration after the first international normalized ratio (INR) monitoring test. The most problematic adverse reaction of VKAs is bleeding, which commonly appears at treatment onset. The risk of bleeding and thromboembolic events is reduced when patients are within the therapeutic range for most of the time [9,10]; therefore, the current clinical goal of VKA therapy is to rapidly achieve this range and maintain patients within it for as long as possible. VKAs have proven to be equally as effective as new oral anticoagulants when patients remain within the therapeutic range for more than 66% of the time [11–13]. In attempts to improve the management of VKA therapy, various researchers have developed algorithms that include clinical and pharmacogenetic parameters for warfarin-anticoagulated patients [1,4,14–21]. However, fewer pharmacogenetic algorithms have been published for ACN [22–27], and only one has been validated externally [23]. Most of these have been developed in White populations, except for one in Indian patients [24]. The main indications for ACN pharmacogenetic algorithms are AF, VTE, and valve replacement, and all models include clinical variables and VKORC1/CYP2C9 DOI: 10.1097/FPC.0000000000000082

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

502 Pharmacogenetics and Genomics 2014, Vol 24 No 10

gene polymorphisms [22–27]. Some authors also explored the additional influence of CYP4F2*3 (rs2108622) [25,26], APOE (rs7412) [25], and GGCX (rs11676382) [24] gene polymorphisms on the stable ACN dose. The percentage variability in the stable ACN dose explained by the different pharmacogenetic algorithms has ranged between 41.4 and 60.6% [22–26], leaving a considerable percentage unexplained. CYP2C19 variants, which present wide interethnic frequencies [28], are among the few other gene polymorphisms considered as potential candidates to explain ACN dose variability. S-ACN, the metabolite responsible for anticoagulant activity, is primarily metabolized by CYP2C9 and 20% by CYP2C19 [29,30]. Differences in CYP2C19 activity in Caucasians can largely be explained by the inactive allele CYP2C19*2 (exon 5) and the increased activity allele CYP2C19*17 [28]. The gene encoding gamma-glutamyl carboxylase (GGCX) has also been studied as a potential candidate gene for this purpose [31–34]. GGCX rs12714145 was found to have a slight effect on the warfarin dose in Swedish patients [31] and was included in an ACN pharmacogenetic algorithm for Indian patients [24]. Apolipoprotein E (APOE) is a polymorphic protein defined by three alleles (E2, E3, and E4) derived from two single nucleotide polymorphisms (rs429358 and rs7412) [35], and it mediates the absorption of vitamin K-rich lipoproteins by hepatic and other tissues [36]. APOE E4-allele carriers were found to require lower ACN doses [37] and higher warfarin doses [38] to achieve stable anticoagulation. The C3435T gene polymorphism of the ABCB1 (adenosine triphosphate-binding cassette) gene has a major impact on the oral bioavailability of numerous drugs transported by P-glycoprotein, which is encoded by this gene [39,40]; thus, the pharmacokinetics of warfarin may be altered in the presence of the ABCB1-C3435T gene polymorphism [41,42]. The influence of P-glycoprotein on ACN distribution and pharmacological action has not yet been examined, although similar pharmacodynamics might be expected, given the structural similarities between ACN and warfarin [43]. Very few studies have been published on the influence of ABCB1 C3435T on VKA dose requirements, and the results have been controversial [41,43,44]. The aim of this study was to develop an ACN pharmacogenetic algorithm for patients with AF or VTE, the most common indications for long-term ACN therapy, by assessing the influence on the stable ACN dose variability of clinical factors, of gene polymorphisms whose association with this variability is well documented [CYP2C9*2 (rs1799853), CYP2C9*3 (rs1057910), VKORC1 (rs9923231), CYP4F2*3 (rs2108622)], and of gene polymorphisms whose association has not yet been fully investigated [ABCB1 C3435T (rs1045642), APOE (rs429358 and rs7412), CYP2C19*2 (rs4244285),

CYP2C19*17 (rs12248560), and GGCX (rs12714145 and rs11676382)].

Patients and methods A retrospective observational study was carried out. Study population

The study included adult patients (≥18 years) diagnosed with AF or VTE and treated with a stable ACN dose at the Complejo Hospitalario de Granada between March and June 2013 after the exclusion of pregnant patients and those with chronic renal failure. The study was approved by the ethics committee of our hospital and complied with the Declaration of Helsinki. All patients signed informed consent to participation in the study. The total cohort was divided randomly into two groups: an algorithm cohort that included around 80% of the patients and a validation cohort that included the remaining patients. Clinical variables

Clinical data were collected by interviews and complemented with clinical records. The clinical variables considered were age (years), sex, BMI (kg/m2), pathology (AF/VTE), CYP2C9 inhibitors (omeprazole/simvastatin/amiodarone), smoking status (yes/no), and alcohol intake (yes/no). Genetic variables

Gene polymorphisms were analyzed by PCR plus restriction fragment length polymorphism [CYP2C9*2 (rs1799853), VKORC1 (rs9923231), CYP4F2 (rs2108622)], direct sequencing [CYP2C9*3 (rs1057910), APOE (rs429358, rs7412), and GGCX (rs12714145)] or real-time PCR using TaqMan probes [ABCB1 C3435T (rs1045642), CYP2C19*2 (rs4244285), CYP2C19*17 (rs12248560), and GGCX (rs11676382)]. The genotyping methodology has been described previously [45]. Response variable

Dose was considered stable when three consecutive INR monitoring test results were within the therapeutic range with a dose variation less than 10% during a 3-month period. Statistical analysis and algorithm construction and evaluation

After assessment of the normality of data distribution using the Kolmogorov–Smirnov test, quantitative data were expressed as mean ± SD for normally distributed variables and as medians and percentiles (25 and 75) for non-normally distributed variables. Student’s t-test or analysis of variance was used for the former and the nonparametric Mann–Whitney U-test or the Friedman test for the latter. The Pearson χ2-test for bivariate association of qualitative variables was used to compare genotype frequencies between algorithm and validation

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 503

cohorts. Analysis of variance or the Kruskal–Wallis test was used to analyze the influence on the dose of independent qualitative variables with more than two categories. Pearson correlation or Spearman tests were used to compare quantitative variables according to the normality of the data distribution (see table footnotes). The Hardy–Weinberg equilibrium was determined using the calculator provided by the Online Encyclopedia for Genetic Epidemiology studies [46]. Construction of the algorithms

Clinical and pharmacogenetic dose algorithms were obtained by multivariate analysis (multiple linear regression) to evaluate the independent influence of each variable on the stable dose requirement and to identify potential confounders. The algorithms were constructed using the data from the algorithm cohort. Variables with P-value up to 0.05 in the bivariate statistical analysis or with special biological significance were included in the multiple linear regression. After obtaining the clinical and pharmacogenetic algorithms, they were tested in the validation cohort to establish their accuracy in predicting the stable ACN dose requirements of the patients as a function of the clinical and pharmacogenetic variables included. Predictive accuracy and clinical significance of the algorithms

The total variability explained by the pharmacogenetic or clinical dosing algorithms was evaluated according to the uncorrected coefficient of determination (R2) obtained from the linear multiple regression analysis. Deviations in the prediction were assessed by calculating the mean error (ME), defined as the mean of the differences between predicted and observed ACN doses. The predictive precision of the models was assessed by calculating the mean absolute error (MAE) as the square root (ME2). These values were also expressed as percentages of the observed ACN dose (%ME and %MAE). Comparisons of the ME and MAE between the algorithms were performed using a paired t-test or a Wilcoxon signed-rank test according to the normality of the variables.

15% higher and greater than 15% lower than the actual dose, respectively. The NNG before observing a significant benefit of using the pharmacogenetic algorithm was computed using the standard number-needed-to-treat method of the International Warfarin Pharmacogenetics Consortium [1], as the inverse of the absolute risk reduction, calculated as the absolute difference between the pharmacogenetic and the clinical algorithms in the rate of predicted dose with deviation from the actual stable dose of greater than 10/15/20%. The predictive performance, deviation, accuracy, and clinical relevance of other published ACN pharmacogenetic algorithms were also tested in our total cohort of patients. ACN pharmacogenetic models were only selected for comparison when the equation for the dose calculation was available and when the parameters included in the original model could be obtained in our cohort of patients. P-value less than 0.05 was considered significant in all tests. SPSS 19.0 software for Windows (IBM plc, Chicago, Illinois, USA) was used for the data analyses.

Results Patient clinical characteristics

The total cohort included 164 patients. As shown in Table 1, there were no differences in the clinical characteristics between the patients in the algorithm (n = 134) and the validation (n = 30) cohorts. The median age of the total cohort was 74 (67–80) years and 54.3% (89/164) were men. The most frequent indication for oral anticoagulation was AF (82.3%; 135/164). Concomitant drugs prescribed with ACN were omeprazole (46.3%; 76/164), simvastatin (34.8%; 57/164), and amiodarone (2.4; 4/164). The median stable dose was 14 (10–18) mg/week. Genotype analysis

All gene polymorphism distributions were in Hardy–Weinberg equilibrium. There were no significant differences in the genotype frequency distribution between the algorithm and the validation cohorts (Table 2). Stable ACN dose requirement as a function of clinical and genetic variables

The potential benefit of the algorithms was compared by calculating the number needed to genotype (NNG), defined as the number of patients who must be genotyped for one patient to have an improved dose estimate, as proposed by The International Warfarin Pharmacogenetics Consortium [1].

The median stable ACN dose was 13 (10–18) mg/week. Older patients required lower stable doses (P = 0.012), with a gradual reduction between the ages of 65 and 80 years (Fig. 1a). The stable dose was also influenced by the BMI (P = 0.037), with a gradual increase in patients with higher BMI (Fig. 1b).

The clinical significance of algorithms was evaluated by calculating the percentage of patients whose predicted ACN dose was within 10–15–20% of the actual stable therapeutic dose. Dose overestimation and underestimation were calculated as the percentage of patients for whom the algorithm-predicted dose was greater than

A trend toward a higher stable dose requirement was observed in patients receiving concomitant simvastatin [13 (9–17) vs. 14 (11–20) mg/week; P = 0.062; Supplementary Table S1, Supplemental digital content 1, http://links.lww. com/FPC/A763]. Although not significant, this variable was considered in the construction of the algorithms to avoid a

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

504

Pharmacogenetics and Genomics 2014, Vol 24 No 10

Table 1

Distribution of clinical variables and comparison between cohorts

Variables

Total cohort

Algorithm cohort

Validation cohort

P (AC vs. VC)

Number of patients Age Sex (women) Indication Atrial fibrillation Venous thromboembolism BMI (kg/m2) Omeprazole Simvastatin Amiodarone Smoking status Alcohol intake Stable dose (mg/week)

164 73 ± 9 75/164 (45.7)

134 73 ± 10 59/134 (44.0)

30 73 ± 8 16/30 (53.3)

0.771 0.355

135/164 (82.3) 29/164 (17.7) 30 (27–33) 76/164 (46.3) 57/164 (34.8) 4/164 (2.4) 6/139 (4.3) 48/139 (34.5) 14 (10–18)

109/134 (81.3) 25/134 (18.7) 29 (27–33) 64/134 (44.8) 45/134(33.6) 4/134 (3.0) 4/111 (3.6) 39/111 (35.1) 13 (10–18)

26/30 (86.7) 4/30 (13.3) 32 ± 5.7 16/30 (53.3) 12/30 (40.0) 0/30 (0.0) 2/28 (7.1) 9/28 (32.1) 14.6 ± 5.0

0.490 0.201 0.396 0.505 0.345 0.410 0.766 0.488

Normally distributed variables: mean ± SD (P: t-paired test). Non-normally distributed variables: median [P25–P75] (P: Mann–Whitney U-test).

potential underestimation of its influence. A similar effect was observed in patients with concomitant amiodarone [9 (8–13) vs. 13 (10–18) mg/week; P = 0.143; Supplementary Table S1, Supplemental digital content 1, http://links.lww. com/FPC/A763], and this variable was also considered. A strong association was found between the VKORC1 genotype and stable dose requirement (Fig. 1c; P = 2 × 10 − 10). The stable ACN dose to achieve stable anticoagulation in CC patients was 27.8% higher than that in CT patients and 61.1% higher than in TT patients (Fig. 1c). The stable dose was associated with the CYP2C9*3 genotype, finding a higher ACN dose (P = 0.001) in wild-type (AA) patients [14 (10–19) mg/week] than in patients with the AC genotype [10 (7–13) mg/week] (Supplementary Table S1, Supplemental digital content 1, http://links.lww. com/FPC/A763). Doses were 21.4% lower for carriers of any allelic variant of the CYP2C9 haplotype (any combination of *2 or *3) than in wild-type (*1/*1) patients (Fig. 1d). The stable dose requirement was nonsignificantly higher in carriers of the T-allele of the CYP4F2 gene polymorphism than in wild-type (CC) patients [14 (10–18) vs. 12 (9–17); P = 0.257; Fig. 1e], but this variable was considered in the construction of the pharmacogenetic algorithm to avoid a potential underestimation of its influence. Finally, the stable ACN dose requirement was 31.6% higher (P = 0.05) in patients with the CC genotype in APOE-rs429358 than in the remaining patients (Fig. 1f). No other clinical or genetic variables were found to influence the stable ACN dose requirement in the algorithm cohort (Supplementary Table S1, Supplemental digital content 1, http://links.lww.com/FPC/A763). Construction of the ACN dosing algorithm according to clinical and genetic variables

Clinical and pharmacogenetic variables with an influence on stable ACN doses in the algorithm cohort in the

bivariate analysis were included in the construction of clinical and pharmacogenetic ACN dosing models. Table 3 shows the variables included in each algorithm (clinical and pharmacogenetic), the percentage of variability explained by each variable, and the equations that predict a stable ACN dose as a function of clinical and genetic variables. The clinical model included age, BMI, and concomitant simvastatin as independent variables and predicted 14% of the variability in the stable ACN dose observed (Table 3). In the pharmacogenetic dosing algorithm, the clinical variables that were associated independently with stable dose were age, BMI, and concomitant amiodarone, which explained 12.1% of the dose variability. Despite showing no significant association in the bivariate analysis, simvastatin and amiodarone were tested in the final clinical and pharmacogenetic algorithms to consider the same clinical factors in both models. The multivariate analysis showed an influence of simvastatin in the clinical algorithm alone and of amiodarone in the pharmacogenetic model alone (Table 3). Besides the clinical variables, the pharmacogenetic model included the CYP2C9 haplotype (*1/*1, *1/*2, *1/*3, *2/*2), VKORC1 genotype (CC, CT, TT), and APOE genotype (CC). The CYP2C9 haplotype, which comprises rs1799853 and rs1057910, predicted 9.8% of the stable dose requirement. The VKORC1 gene polymorphism was the most potent predictor variable, explaining 31.5% of the total variability in stable ACN dose requirement, whereas the CC genotype for the APOE-rs429358 haplotype explained 3.9%. The pharmacogenetic model explained 56.6% of the variability in the stable ACN dose in the algorithm cohort.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 505

Table 2

Genotype distribution between algorithm and validation

cohorts Variables

Algorithm cohort (N = 134)

CYP2C9*3rs1057910 AA 116 (86.6) AC 18 (13.4) CC 0 (0.0) CYP2C9*2 rs1799853 CC 97 (72.4) CT 33 (24.6) TT 4 (3.0) VKORC1*3 rs9923231 CC 48 (35.8) CT 60 (44.8) TT 26 (19.4) CYP4F2*3 rs2108622 CC 42 (31.6) CT 76 (57.1) TT 15 (11.3) ABCB1 C3435T rs1045642 CC 36 (26.9) CT 67 (50.0) TT 31 (23.1) APOE rs429358 CC 9 (6.7) CT 24 (17.9) TT 101 (75.4) APOE rs7412 CC 123 (91.8) CT 11 (8.2) TT – CYP2C19*2 rs4244285 GG 84 (62.7) AG 32 (23.9) AA 18 (13.4) CYP2C19*17 rs12248560 CC 73 (55.7) CT 48 (36.6) TT 10 (7.6) GGCX rs12714145 CC 39 (29.3) CT 69 (51.9) TT 25 (18.8) GGCX rs11676382 CC 113 (84.3) CG 21 (15.7) GG – CYP2C9 haplotype *1/*1 (WT) 82 (61.2) *1/*2 30 (22.4) *1/*3 15 (11.2) *2/*2 4 (3.0) *2/*3 3 (2.2) *3/*3 0 (0.0) APOE haplotype E2/E3 11 (8.2) E3/E3 90 (67.2) E3/E4 24 (17.9) E4/E4 9 (6.7)

Validation cohort (N = 30)

P

26 (86.7) 3 (10.0) 1 (0.0)

0.513

23 (76.7) 6 (20.0) 1 (3.3)

0.710

11 (36.7) 14 (46.7) 5 (16.7)

0.941

13 (44.8) 14 (48.3) 2 (6.9)

0.169

10 (33.3) 15 (50.0) 5 (16.7)

0.659

1 (3.3) 5 (16.7) 24 (80.0)

0.759

27 (90.0) 3 (10.0) –

0.722

19 (63.3) 5 (16.7) 6 (20.0)

0.525

21 (70.0) 5 (16.7) 4 (13.3)

0.094

9 (30) 18 (60) 3 (10)

0.5296

24 (82.8) 5 (17.2) –

0.785

19 6 3 1

(63.3) (20.0) (10.0) (3.3) 0 1 (3.3)

3 21 5 1

(10.0) (70.0) (16.7) (3.3)

cohort, and 0.2% in the validation cohort, whereas the pharmacogenetic algorithm explained 56.6% of the variability in the total cohort, 56.6% in the algorithm cohort, and 57.3% in the validation cohort (Fig. 2). Predictive accuracy and clinical significance

–

–

P: Pearson’s χ2-test.

Performance of clinical and pharmacogenetic algorithms

The ACN dosing algorithms were tested in the validation cohort (n = 30 patients), the algorithm cohort (n = 134 patients), and the total cohort (n = 164 patients). The performance of each algorithm was comparable between the total cohort and the algorithm cohorts, yielding a better prediction of the stable ACN dose with the pharmacogenetic algorithm (Fig. 2). The clinical algorithm explained only 10.9% of the stable ACN dose variability in the total cohort, 14.0% in the algorithm

Both clinical and pharmacogenetic algorithms showed a very low deviation, but a wide variability when expressed as ME and %ME (Table 4). However, the absolute deviation from the stable dose, expressed as MAE, was ≈1.5-fold lower for the pharmacogenetic algorithm, which evidenced greater accuracy in all three cohorts; thus, the median %MAE for the clinical algorithm in the validation cohort was 1.7-fold higher than that of the pharmacogenetic algorithm (Table 4). In addition, doses predicted by the pharmacogenetic algorithm showed lower variability as can be observed by the much narrower interquartile ranges (Table 4). The clinical relevance of the algorithms was evaluated by calculating the percentage of patients whose predicted ACN dose was within 10–15–20% of the actual stable therapeutic dose (Fig. 3a). In the total cohort, the pharmacogenetic algorithm provided a more correct classification of the patients in comparison with the clinical algorithm, even when the deviation from the stable dose was as low as 10%. In the algorithm cohort, 29% of patients were correctly classified by the pharmacogenetic algorithm with a deviation of up to 10%, whereas the clinical algorithm only reached this percentage when a deviation of 20% was considered. In the validation cohort, 53.3% correct classification (≤ 15%) was achieved with the pharmacogenetic algorithm versus 27% with the clinical model (Fig. 3a). Out of the 46.7% (14/30) patients classified incorrectly with the pharmacogenetic algorithm, 92.3% (13/14) received an intermediate dose (7–21 mg/week) and one received an elevated dose (>21 mg/week). Dose overestimations (>15%) and underestimations (< 15%) were lower for the pharmacogenetic versus the clinical algorithm in all cases (Fig. 3b). The NNG in the total cohort was 5.9 (4–16) for a 20% deviation from the stable dose and increased to ≈7.5 for deviations of 10 and 15% (Fig. 3a). The absolute risk reduction was 17.1% greater with the pharmacogenetic versus the clinical algorithm. Comparison of performance, predictive accuracy, and clinical significance of other ACN pharmacogenetic algorithms

Three ACN dosing pharmacogenetic algorithms available in the literature were tested in the total cohort of patients (n = 164 patients) for comparative purposes [22,23,25]. Table 5 shows data on their performance, predictive accuracy, and clinical significance. The Markatos algorithm, developed in oral anticoagulated patients, included age and CYP2C9 and VKORC1 gene polymorphisms, and explained 55% of the total ACN stable dose

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

506 Pharmacogenetics and Genomics 2014, Vol 24 No 10

Fig. 1

(a)

Dose (mg/week)

25

P =0.012

20

20

15

15

10

10

5

r2 lineal = 0.136

P =0.037

5

r2 lineal = 0.057

0

0 40

50

60

70 Age

80

90

100

20

30 BMI

30

30

25

25

20

144 127

15 10

50

20 15 10

∗

5

5 #

#

∗

#

#

∗1/∗1

∗1/∗2

∗1/∗3

0

0 CC

CT

TT

VK0RC1

∗2/∗2

∗2/∗3

CYP2C9

(e)

(f) 30

Stable dose (mg/week)

40

(d)

(c)

Stable dose (mg/week)

(b) 25

62

30

25

25

20

20

15

15

10

10

5

5

0

0 T-allele

CC CYP4F2

125 P =0.05

T-allele

CC

APOE-rs429358

Stable acenocoumarol dose as a function of clinical and pharmacogenetic variables in the algorithm cohort. (a) Stable acenocoumarol dose required as a function of age. (b) Stable acenocoumarol dose required as a function of BMI. (c) Stable acenocoumarol dose required as a function of the VKORC1 gene polymorphism. *Different from the other two categories (CC-CT: P = 0.0001; CC-TT: P = 9 × 10 − 12); #Different from each other (CTTT: 2 × 10 − 7). (d) Stable acenocoumarol dose required as a function of the CYP2C9 haplotype. *Different from the other two categories (*1/*1 vs. *1/*3, P = 0.0003; *1/*1 vs. *2/*3, P = 0.017); #Different from each other (*1/*2 vs. *1/*3, P = 0.022). (e) Stable acenocoumarol dose required as a function of the CYP4F2 gene polymorphism. (f) Stable acenocoumarol dose required as a function of the APOE-rs429358 gene polymorphism.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 507

Table 3

Construction of acenocoumarol dosing algorithms as a function of clinical and genetic variables

Variables

Clinical algorithm

Pharmacogenetic algorithm

P-value

Cumulative R2 (%)

15.470 − 0.145 0.276

7.962 − 0.116 0.286

0.002 0.0003

6.2 4.4

–

0.016

2.8

− 5.295

0.014

1.5 9.8

Constant Age (years) BMI (kg/m2) Simvastatin Yes Amiodarone Yes CYP2C9 haplotype (WT)*1/*1 *1/*2 *1/*3 *2/*2 VKORC1-rs9923231 CT TT APOE-rs429358 CC Unadjusted R2 of the algorithm (%)

2.506 –

−5

– – – –

10.596 10.586 5.875 8.387

3 × 10 6 × 10 − 5 0.026 0.009

– –

− 4.748 − 9.388

5 × 10 − 8 3 × 10 − 16

– 14.0

3.742 56.6

0.011

31.5

3.9

Fig. 2

PGx algorithm Validation cohort

Clinical algorithm

PGx algorithm Algorithm cohort

Clinical algorithm

PGx algorithm Total cohort

Clinical algorithm 0%

20%

40%

60%

80%

100%

Percentage stable dose variability explained by clinical and pharmacogenetic algorithms. Black bars, explained percentage; Grey bars, unexplained percentage.

Table 4

Predictive performance of clinical and pharmacogenetic algorithms Bias

Cohort

Algorithm

Total (N = 164)

Clinical PGx Clinical PGx Clinical PGx

Algorithm (N = 134) Validation (N = 30)

ME (mg/week)

P

0.018 ± 5.5 − 0.01 ± 3.8 0.3 ± 5.5 5 × 10 − 4 ± 3.8 0.04 ± 5.5 0.5 [ − 1.7, 2.5]

0.921 0.925 0.971

Accuracy %ME 1.5 1.1 3.7 0.4 − 9.4 − 3.2

[ − 20, [ − 15, [ − 23, [ − 16, [ − 19, [ − 10,

P 38] 21] 39] 27] 40] 15]

0.882 0.833 9 × 10 − 5

MAE (mg/week) 3.8 2.5 4.0 2.7 3.0 1.9

[2, 7] [1, 4] [2, 7] [1, 4] [3, 7] [0.7, 4]

P 2 × 10 − 9 4 × 10 − 7 0.001

MAE (%) 26.8 17.7 26.8 18.1 23.3 13.5

[13, 46] [8, 34] [13, 47] [8, 36] [14, 46] [8, 28]

P 5 × 10 − 8 1 × 10 − 6 0.011

MAE, mean absolute error [SQR(ME2]; ME, mean error in stable dose (predicted dose − observed dose); PGx, pharmacogenetic. Normally distributed variables: mean ± SD (P: paired t-test). Non-normally distributed variables: median [P25, P75] (P: Wilcoxon signed-rank test).

variability in their population [22]. The EU-PACT pharmacogenetic dosing algorithm considered age, female sex, height, weight, and amiodarone as clinical variables and CYP2C9 and VKORC1 as genetic variables; it explained 52.6% of the total variability in their

derivation cohort and obtained similar results in a different (external) Dutch population [23,47]. The pharmacogenetic algorithm published by Borobia et al. [25], which was developed in 117 VTE patients and included sex, BMI, and CYP2C9 inhibitors/inducers as clinical

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

508

Pharmacogenetics and Genomics 2014, Vol 24 No 10

Fig. 3

Clinical algorithm Pharmacogenetic algorithm P =0.03

P =0.005

P =0.04

P =0.002

60

P =0.001

P =0.009

80

P =0.003

P =0.001

100

P =0.148

(a)

67 55

53

52 50

40 40 31

40

29

27

20 17 0

17

TC

28

VC

TC

AC

≤10%

11.9

23.3

[4−21]

AC

ARR (Cl95%)

9.7

26.7

[3−23]

[2−21]

[2−47]

17.1

[5−55] [2.3−60]

17.2

16.7

[6−27] [5−28] [8−39]

NNG (Cl95%) 4.3

VC

≤20%

13.3

NNG (Cl95%) 8.4

TC

ARR (Cl95%)

[2−22] [0.5−43]

7.1

VC

≤15%

ARR (Cl95%)

[5−23]

35

27

17

AC

14.2

38

37

NNG (Cl95%)

7.8

10.3

3.8

[4−39]

[5−66]

[2−52]

5.9

5.8

6.0

[4−16] [4−19] [3−13]

(b) VC-clinical

27.0

37 17

VC-PGx

31

AC-PGx

28

TC-clinical

36

27

32 25 10

42

37

27

0

30

53.0

AC-clinical

TC-PGx

37

41

40 20

30

40

35 50

60

70

80

90

100

Clinical significance of algorithms. (a) Patients correctly classified by each algorithm as a function of deviations of 10, 15, or 20% from the stable acenocoumarol dose. P-value: McNemar test for paired proportions. Absolute risk reduction (ARR) and number needed to genotype (NNG) in each cohort. (b) Overestimation (dark gray) and underestimation (light gray) of doses (≤ 15%) by each algorithm in the three cohorts. Black: percentage of correctly classified patients. CI, confidence interval.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 509

Table 5

Predictive performance, deviation, accuracy, and clinical significance of pharmacogenetic algorithms in the total cohort Deviation

Cohort

Algorithm

Total (N = 164) Markatos Borobia EU-PACT This study

2

R (%) ME (mg/week) 21.4 42.2 53.1 56.6

− 0.6 [ − 5, 3] 2.1 [ − 2, 5] 1.6 ± 4.0 − 0.01 ± 3.8

P

Accuracy

%ME −5

9 × 10 10 − 9 3 × 10 − 17

−4 19 15 1.4

P

[ − 29, 22] 0.917 [ − 9, 47] 2 × 10 − 9 [ − 6, 39] 2 × 10 − 12 [ − 15, 25]

MAE (mg/week) 3.9 3.6 3.1 2.5

[2, [2, [2, [1,

6] 6] 5] 4]

P

MAE (%) −6

2 × 10 3 × 10 − 6 0.002

28 25 22 17.7

P

[14, 42] 4 × 10 − 5 [12, 48] 4 × 10 − 8 [11, 39] 3 × 10 − 5 [8, 34]

Correct classification (≤ 20%) 37.8 39.3 43.9 55

MAE, mean absolute error [SQR(ME2)]; ME, mean error in stable dose (predicted dose − observed dose); PGx, pharmacogenetic; R2, Spearman correlation. Normally distributed variables: mean ± SD (P: paired t-test). Non-normally distributed variables: median [P25, P75] (P: Wilcoxon signed-rank test).

variables and CYP2C9, VKORC1, CYP4F2, and APOErs7412 as genetic variables, explained 60.6% of the total dose variability in their derivation cohort. Application of these pharmacogenetic algorithms in our cohort of patients explained between 21.4 and 53.1% of the total ACN dosage variability (Table 5). All of these pharmacogenetic algorithms showed a higher deviation and a lower accuracy in our total cohort in comparison with the present pharmacogenetic model, which was 24% more accurate in the prediction of the stable ACN dose in comparison with the external model showing the greatest accuracy (MAE and %MAE; P = 0.002 and 3 × 10 − 5, respectively) (Table 5). The percentage correct classification (with deviation of ≤ 20%) of the EU-PACT pharmacogenetic model (43.9%) was higher than that obtained with the other models, but was 22% lower than that obtained with our proposed algorithm.

Discussion In long-term anticoagulant therapy, the initial phase of VKA treatment is the most challenging period because the dose needed to achieve stable anticoagulation and maintain the therapeutic INR is unknown; therefore, numerous empirical dose adjustments are required [47], exposing patients to a higher risk of thromboembolic (INR < 2) and bleeding events (INR > 4) [48,49]. In the present study, the influence of clinical and pharmacogenetic parameters on the stable ACN dose requirement was investigated to develop a pharmacogenetic algorithm comprised of clinical and genetic variables for a more accurate prediction of the stable ACN dose in AF and VTE patients. The goal is to implement pharmacogenetic algorithms in the clinical management of patients initiating ACN therapy to minimize the duration of the first dose titration phase and maximize the TTR, reducing the risk of bleeding and thromboembolic events for as long as possible. Construction of the pharmacogenetic algorithm was performed on the basis of data from 134 patients with a stable ACN dose and tested in 30 patients with similar characteristics. Of the clinical and genetic variables tested for the algorithm, only age, BMI, concomitant amiodarone, and CYP2C9 (*1/*1;*1/*2;*1/*3;*2/*2), VKORC1

(CT; TT), and APOE-rs429358 (CC) gene polymorphisms contributed toward the final algorithm equation. Older age and concomitant amiodarone were associated with lower stable ACN dose and greater BMI with a higher dose, as reported in previous pharmacogenetic algorithms, but together accounted for only 12.1% of the dose variability [23–26]. Concomitant simvastatin or omeprazole did not contribute toward the final model, as observed previously [22,26,27]. It has been reported widely that the main gene polymorphisms related to stable ACN dose variability are VKORC1, CYP2C9, and, to a lesser extent, CYP4F2. Consistent with these findings, the VKORC1-rs9923231 gene polymorphism and the CYP2C9 haplotype were responsible for most of the total variability in stable ACN dose in the present pharmacogenetic algorithm. The CYP4F2 V433M (rs2108622) gene polymorphism was not finally entered into our pharmacogenetic algorithm, despite previous reports of its association with a higher VKA dose requirement [50,51] and its inclusion in other pharmacogenetic algorithms [25,26]. Our finding is consistent with published evidence of the very low influence of this gene polymorphism on ACN dose variability (1–2%) [26,51,52]. Few studies have evaluated the influence of APOE polymorphisms on ACN dose requirements, and their results have been inconsistent [2,25,37,38]. In our series, 3.9% of the variability was explained by the APOErs429358-CC genotype. The stable ACN dose was 31.6% higher in patients with this genotype than in the remaining patients, in agreement with findings in warfarin-anticoagulated patients [38]. We recently showed that APOE E3/E3 patients, who are TT for the APOE-rs429358 gene polymorphism, have an increased risk of overanticoagulation (INR > 4) in the initial phase of ACN treatment [45]. The present study is the first to achieve optimization of the dose prediction by adding this gene polymorphism to a pharmacogenetic algorithm. The APOE-rs7412 gene polymorphism, which showed no influence in our patients, was included in a previous algorithm [25], being associated with a higher stable dose in VTE patients (P = 0.067). The noninclusion of APOErs429358-CC in the earlier study may be attributable to

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

510 Pharmacogenetics and Genomics 2014, Vol 24 No 10

the significantly lower frequency of patients with this polymorphism (1.7%) in comparison with the present study (6.7%) [25]. The influence of ABCB1 C3435T on VKAs dose requirements is controversial [41,43,44]. The ABCB1–3435CC genotype has been associated with a lower VKA dose [43,44], whereas the warfarin dose was found to be 24% lower in patients heterozygous for ABCB1 haplotype D, which includes C3435T nucleotide exchange [41]. In our patients, the ABCB1 gene polymorphism showed no influence on stable ACN dose and was not included in the final model. No influence on ACN dose variability was observed for the CYP2C19 gene polymorphism, as observed previously [43,50], or for GGCX rs12714145 or rs11676382 gene polymorphisms. GGCX polymorphisms have been included in only two VKA pharmacogenetic algorithms, which were not developed in White patients [24,51]. According to our findings [CYP4F2*3 (rs2108622), ABCB1 C3435T (rs1045642), APOE (rs7412), CYP2C19*2 (rs4244285), CYP2C19*17 (rs12248560), and GGCX (rs12714145 and rs11676382)] gene polymorphisms have no influence on the prediction of stable ACN dose and appear not to be relevant for the optimization of ACN dose algorithms. Over the past 4 years, several algorithms combining clinical and pharmacogenetic factors have been developed to improve ACN dosage in the initiation of anticoagulation therapy, on the basis of different patients and initiation protocols, but with similar findings (Table 6). The EUPACT genotype dosing algorithm was validated externally in 168 ACN-treated patients and explained 47.3% of the variability in maintenance dose [23]. More recently, two pharmacogenetic algorithms were developed in Spanish patients, and both included genetic variations in VKORC1, CYP2C9, and CYP4F2 [25,26]; the one that explained the highest percentage of ACN dose variability (60.6%) was developed in 117 VTE patients and also included APOErs7412 [25]. A better algorithm performance has not been achieved using larger sample sizes, given that the algorithm developed in the largest cohort of patients published to date (3656 anticoagulated patients with a derivation cohort of 973 patients) could not explain more than 50% of the variability in stable dose [26]. Our pharmacogenetic algorithm explained 56.6% of the dose variability in the total cohort, the highest percentage published to date in AF and VTE patients. Unlike in other studies, a similar performance was obtained in the validation cohort (57.3% of dose variability). An additional 3.9% of dose variability was explained by including a new gene polymorphism of APOE, but not by incorporating other potential gene candidates that have been little studied, such as CYP4F2, CYP2C19, ABCB1, and GGCX. Novel polymorphisms of VKORC1, that is, rs61742245 and rs55894764 variants [52–55], have been associated with a slightly higher stable ACN requirement

and produced an improvement of ≈1% in one prediction model [52]. Given that the loss of rs9923231/rs9934438 linkage has been described in some patients, the determination of rs9934438 may provide a better dose adjustment in these cases [52]. The novel CYP2C9*57 haplotype, which was detected in one out of 309 oral anticoagulated patients (0.3%), showed VKA hypersensitivity because of a slower warfarin and ACN metabolism, mandating a dose reduction, and it was associated with multiple bleeding episodes and an increased percentage of INRs above 3.0 [56]. The patient with the novel CYP2C9*57 haplotype was also wild type for the most common gene polymorphisms of VKORC1 and CYP2C9, which may have masked his true dose requirement [56]. Other variants associated with ACN dosage variation in a genome-wide study were CYP2C9 (rs4086116) and CYP2C18 (rs1998591, rs12772169, rs1042194) [57], which have recently shown frequencies at least 0.8% in various population samples [58]. All of these gene variants and other gene alterations, such as copy number variation or exome sequencing, may evidence novel associations with ACN dose requirements that could reduce the percentage of variability that is not explained by current pharmacogenetic dosing algorithms. The absolute deviation from the stable ACN dose (MAE) yielded by our pharmacogenetic algorithm in the validation cohort was 50% lower than that in previous reports (1.9 vs. 3.75–4.41 mg/week) [23,25,27]. In fact, 67% of patients in our validation cohort were correctly classified (with deviation of ≤ 20%) by the pharmacogenetic algorithm, in comparison with the 40–50% obtained with other models, which only reach 60% for some dose ranges [25,26]. Moreover, a similar percentage of our patients in the validation cohort (33%) were correctly classified with a deviation of up to 10%, indicating a higher accuracy in the dose prediction. The percentage correct classification with our model was surpassed by that (70%) obtained with a pharmacogenetic model developed in a Polish population, but only when a deviation of up to 25% from the real dose was considered [27]. Our proposed pharmacogenetic algorithm correctly classified (deviation of ≤ 15%) 53% of patients in the validation cohort, that is, ≈30% more in comparison with the clinical algorithm constructed from routinely gathered variables (P = 0.03). Although the algorithm failed to predict a correct dose (with ≤ 15%) for the remaining 47% of patients, the latter were almost all (92.3%; 13/14) receiving intermediate doses (7–21 mg/week) and therefore may be considered at a lower risk in comparison with patients on more extreme doses [16]. Consequently, the model provided a good adjustment for patients with the more problematic extreme doses. The NNG to avoid misclassifying one patient (≤20%) was ≈6, similar to other pharmacogenetic algorithms [1,25]. In summary, our pharmacogenetic model showed a lower deviation and higher accuracy in the prediction and correct classification of patients,

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Age, BMI, amiodarone, VKORC1, CYP2C9, APOErs429358

We also tested the other ACN pharmacogenetic algorithms available [22,23,25] in our total cohort of patients to evaluate the need for our new pharmacogenetic model in AF and VTE patients. Comparison of the predictive performance, deviation, accuracy, and clinical significance of the other algorithms in our total cohort showed that the correct classification of the initial ACN dosage was significantly lower (by at least 22%) with the other pharmacogenetic models than with our proposed algorithm. The dose recommendation would have deviated by over 20% in more than 56% of the patients if these other algorithms had been applied in our population.

1.9 [0.7, 3.6] mg/week

AF, atrial fibrillation; MAE, mean absolute error; HVR, heart valve replacement; VTE, venous thromboembolism. No height or weight data.

57.3% 56.6% 30 (18.29) 134 (81.7) AF, VTE White This study

973 Anticoagulated patients White

a

48% 50% 2683

226 AF, VTE, HVR White

Wolkanin-Bartnik et al. [27] Cerezo-Manchado et al. [26]

50

surpassing the performance of pharmacogenetic algorithms developed in much larger patient cohorts, suggesting that our smaller sample size was not a limitation.

≤ 20% < 7 mg/week: ≈60% 7–25 mg/week: 40% > 25 mg/week: ≈52% ≤ 10%: 40% ≤ 15%: 53% ≤ 20%: 67%

In ≤ 25%: 4.4 [3.7, 5.0] mg/week – ≤ 25%: 70% – 49%

3.75 mg/week ≤ 20%: 46.7% 38.8% 60.6% 30 (20.40) 117 (79.59) VTE White Borobia et al. [25]

100 (44.44) 41.4% HVR Indian Rathore et al. [24]

AF, VTE, HVR, other AF, VTE, HVR, other

98 375

– 168 White White Markatos et al. [22] EU-PACT [23]

125 (55.5)

–

–

Variables

–

– 3.99 mg/week – – – 47.3%a 54.3 52.6%

MAE (validation cohort) Patients correctly classified Validation Model Indication

Model

Validation Population References

R2 N or [N (%)]

Comparison of construction, performance, predictive accuracy, and clinical significance of acenocoumarol algorithms Table 6

Age, CYP2C9*2/*3, VKORC1 CYP2C9, VKORC1, age, female, height, weight, amiodarone Smoker, male, age, height, weight, BSA, VKORC1, CYP2C9, CYP4F2, GGCX-rs11676382 Age, BMI, other concomitant drugs, VKORC1, CYP2C9, CYP4F2, APOE-rs7412 Age, BMI, vitamin K intake, creatinine clearance < 40 ml/ min, VKORC1, CYP2C9 Age, BMI, body surface area, VKORC1, CYP2C9, CYP4F2

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 511

The benefits of implementing ACN pharmacogenetic algorithms in clinical practice are currently under research [59,60]. The results of only one randomized trial of genotype-guided ACN dosing (EU-PACT group) found that genotype-guided dosing did not improve the percentage of time within the therapeutic range during the first 12 weeks of therapy [60]. However, the validity of pharmacogenetic algorithms in populations other than those used for their development must be established before their implementation in different settings. The present proposal has been tested in a small group of patients (validation cohort) with characteristics similar to those of the patients used for its construction, and further external validation studies are warranted. The actual clinical benefit of its implementation also remains to be elucidated in a clinical trial. In conclusion, we developed a pharmacogenetic algorithm composed of clinical (age, BMI, and concomitant amiodarone) and genetic (CYP2C9, VKORC1, and APOE) factors that can explain 56.6% of the variability in a stable ACN dose, the highest value reported in AF and VTE to date, and can correctly classify 67% of patients with a deviation of up to 20%.

Acknowledgements The authors would like to acknowledge all the professionals from the Virgen de las Nieves University Hospital who contributed toward the management of the samples, especially the nursing team from the Hematology Department. They are also grateful to Antonia Moreno Casares for operating the ABI PRISM 3130xl Genetic Analyzer instrument. The results of this investigation are part of the doctoral thesis presented by Enrique Jiménez-Varo. This work was partly supported by a contract for Marisa Cañadas-Garre (Técnicos de Apoyo Subprogram) from Instituto de Salud Carlos III, Ministerio de Economía y Competitividad.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

512 Pharmacogenetics and Genomics 2014, Vol 24 No 10

Conflicts of interest

21

There are no conflicts of interest.

References 1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

Klein TE, Altman RB, Eriksson N, Gage BF, Kimmel SE, Lee MT, et al. International Warfarin Pharmacogenetics Consortium. Estimation of the warfarin dose with clinical and pharmacogenetic data. N Engl J Med 2009; 360:753–764. Cadamuro J, Dieplinger B, Felder T, Kedenko I, Mueller T, Haltmayer M, et al. Genetic determinants of acenocoumarol and phenprocoumon maintenance dose requirements. Eur J Clin Pharmacol 2010; 66:253–260. Johnson JA, Gong L, Whirl-Carrillo M, Gage BF, Scott SA, Stein CM, et al. Clinical Pharmacogenetics Implementation Consortium. Clinical Pharmacogenetics Implementation Consortium Guidelines for CYP2C9 and VKORC1 genotypes and warfarin dosing. Clin Pharmacol Ther 2011; 90:625–629. Gage BF, Eby C, Johnson JA, Deych E, Rieder MJ, Ridker PM, et al. Use of pharmacogenetic and clinical factors to predict the therapeutic dose of warfarin. Clin Pharmacol Ther 2008; 84:326–331. Limdi NA, Wadelius M, Cavallari L, Eriksson N, Crawford DC, Lee MT, et al. International Warfarin Pharmacogenetics Consortium. Warfarin pharmacogenetics: a single VKORC1 polymorphism is predictive of dose across 3 racial groups. Blood 2010; 115:3827–3834. Schalekamp T, Brassé BP, Roijers JF, van Meegen E, van der Meer FJ, van Wijk EM, et al. VKORC1 and CYP2C9 genotypes and phenprocoumon anticoagulation status: interaction between both genotypes affects dose requirement. Clin Pharmacol Ther 2007; 81:185–193. Rieder MJ, Reiner AP, Gage BF, Nickerson DA, Eby CS, McLeod HL, et al. Effect of VKORC1 haplotypes on transcriptional regulation and warfarin dose. N Engl J Med 2005; 352:2285–2293. Li C, Schwarz UI, Ritchie MD, Roden DM, Stein CM, Kurnik D. Relative contribution of CYP2C9 and VKORC1 genotypes and early INR response to the prediction of warfarin sensitivity during initiation of therapy. Blood 2009; 113:3925–3930. Oake N, Jennings A, Forster AJ, Fergusson D, Doucette S, van Walraven C. Anticoagulation intensity and outcomes among patients prescribed oral anticoagulant therapy: a systematic review and meta-analysis. CMAJ 2008; 179:235–244. Rose AJ, Berlowitz DR, Ash AS, Ozonoff A, Hylek EM, Goldhaber-Fiebert JD. The business case for quality improvement: oral anticoagulation for atrial fibrillation. Circ Cardiovasc Qual Outcomes 2011; 4:416–424. Connolly SJ, Ezekowitz MD, Yusuf S, Eikelboom J, Oldgren J, Parekh A, et al. RE-LY Steering Committee and Investigators. Dabigatran versus warfarin in patients with atrial fibrillation. N Engl J Med 2009; 361:1139–1151. Granger CB, Alexander JH, McMurray JJ, Lopes RD, Hylek EM, Hanna M, et al. ARISTOTLE Committees and Investigators. Apixaban versus warfarin in patients with atrial fibrillation. N Engl J Med 2011; 365:981–992. Patel MR, Mahaffey KW, Garg J, Pan G, Singer DE, Hacke W, et al. ROCKET AF Investigators. Rivaroxaban versus warfarin in nonvalvular atrial fibrillation. N Engl J Med 2011; 365:883–891. Gage BF, Eby C, Milligan PE, Banet GA, Duncan JR, McLeod HL. Use of pharmacogenetics and clinical factors to predict the maintenance dose of warfarin. Thromb Haemost 2004; 91:87–94. Hillman MA, Wilke RA, Caldwell MD, Berg RL, Glurich I, Burmester JK. Relative impact of covariates in prescribing warfarin according to CYP2C9 genotype. Pharmacogenetics 2004; 14:539–547. Kamali F, Khan TI, King BP, Frearson R, Kesteven P, Wood P, et al. Contribution of age, body size, and CYP2C9 genotype to anticoagulant response to warfarin. Clin Pharmacol Ther 2004; 75:204–212. Sconce EA, Khan TI, Wynne HA, Avery P, Monkhouse L, King BP, et al. The impact of CYP2C9 and VKORC1 genetic polymorphism and patient characteristics upon warfarin dose requirements: proposal for a new dosing regimen. Blood 2005; 106:2329–2333. Carlquist JF, Horne BD, Muhlestein JB, Lappé DL, Whiting BM, Kolek MJ, et al. Genotypes of the cytochrome p450 isoform, CYP2C9, and the vitamin K epoxide reductase complex subunit 1 conjointly determine stable warfarin dose: a prospective study. J Thromb Thrombolysis 2006; 22:191–197. Wu AH, Wang P, Smith A, Haller C, Drake K, Linder M, Valdes R Jr. Dosing algorithm for warfarin using CYP2C9 and VKORC1 genotyping from a multiethnic population: comparison with other equations. Pharmacogenomics 2008; 9:169–178. Gong IY, Tirona RG, Schwarz UI, Crown N, Dresser GK, Larue S, et al. Prospective evaluation of a pharmacogenetics-guided warfarin loading and maintenance dose regimen for initiation of therapy. Blood 2011; 118:3163–3171.

22

23

24

25

26

27

28

29

30

31

32

33

34

35 36 37

38

39

40

41

42

Wells PS, Majeed H, Kassem S, Langlois N, Gin B, Clermont J, Taljaard M. A regression model to predict warfarin dose from clinical variables and polymorphisms in CYP2C9, CYP4F2, and VKORC1: derivation in a sample with predominantly a history of venous thromboembolism. Thromb Res 2010; 125:e259–e264. Markatos CN, Grouzi E, Politou M, Gialeraki A, Merkouri E, Panagou I, et al. VKORC1 and CYP2C9 allelic variants influence acenocoumarol dose requirements in Greek patients. Pharmacogenomics 2008; 9:1631–1638. Van Schie RM, Wessels JA, le Cessie S, de Boer A, Schalekamp T, van der Meer FJ, et al. EU-PACT Study Group. Loading and maintenance dose algorithms for phenprocoumon and acenocoumarol using patient characteristics and pharmacogenetic data. Eur Heart J 2011; 32:1909–1917. Rathore SS, Agarwal SK, Pande S, Singh SK, Mittal T, Mittal B. Therapeutic dosing of acenocoumarol: proposal of a population specific pharmacogenetic dosing algorithm and its validation in north Indians. PLoS One 2012; 7:e37844. Borobia AM, Lubomirov R, Ramírez E, Lorenzo A, Campos A, MuñozRomo R, et al. An acenocoumarol dosing algorithm using clinical and pharmacogenetic data in Spanish patients with thromboembolic disease. PLoS One 2012; 7:e41360. Cerezo-Manchado JJ, Rosafalco M, Antón AI, Pérez-Andreu V, GarciaBarberá N, Martinez AB, et al. Creating a genotype-based dosing algorithm for acenocoumarol steady dose. Thromb Haemost 2013; 109:146–153. Wolkanin-Bartnik J, Pogorzelska H, Szperl M, Bartnik A, Koziarek J, Bilinska ZT. Impact of genetic and clinical factors on dose requirements and quality of anticoagulation therapy in Polish patients receiving acenocoumarol: dosing calculation algorithm. Pharmacogenet Genomics 2013; 23:611–618. Furuta T, Graham DY. Pharmacologic aspects of eradication therapy for Helicobacter pylori infection. Gastroenterol Clin North Am 2010; 39:465–480. Godbillon J, Richard J, Gerardin A, Meinertz T, Kasper W, Jähnchen E. Pharmacokinetics of the enantiomers of acenocoumarol in man. Br J Clin Pharmacol 1981; 12:621–629. Teichert M, van Noord C, Uitterlinden AG, Hofman A, Buhre PN, De Smet PA, et al. Proton pump inhibitors and the risk of overanticoagulation during acenocoumarol maintenance treatment. Br J Haematol 2011; 153:379–385. Wadelius M, Chen LY, Downes K, Ghori J, Hunt S, Eriksson N, et al. Common VKORC1 and GGCX polymorphisms associated with warfarin dose. Pharmacogenomics J 2005; 5:262–270. Wadelius M, Chen LY, Eriksson N, Bumpstead S, Ghori J, Wadelius C, et al. Association of warfarin dose with genes involved in its action and metabolism. Hum Genet 2007; 121:23–34. Rieder MJ, Reiner AP, Rettie AE. Gamma-glutamyl carboxylase (GGCX) tagSNPs have limited utility for predicting warfarin maintenance dose. J Thromb Haemost 2007; 5:2227–2234. Pérez-Andreu V, Roldán V, López-Fernández MF, Antón AI, Alberca I, Corral J, et al. Pharmacogenetics of acenocoumarol in patients with extreme dose requirements. J Thromb Haemost 2010; 8:1012–1017. Zannis VI, Just PW, Breslow JL. Human apolipoprotein E isoprotein subclasses are genetically determined. Am J Hum Genet 1981; 33:11–24. Shearer MJ. Vitamin K metabolism and nutriture. Blood Rev 1992; 6:92–104. Visser LE, Trienekens PH, De Smet PA, Vulto AG, Hofman A, van Duijn CM, Stricker BH. Patients with an ApoE epsilon4 allele require lower doses of coumarin anticoagulants. Pharmacogenet Genomics 2005; 15:69–74. Kimmel SE, Christie J, Kealey C, Chen Z, Price M, Thorn CF, et al. Apolipoprotein E genotype and warfarin dosing among Caucasians and African Americans. Pharmacogenomics J 2008; 8:53–60. Hoffmeyer S, Burk O, von Richter O, Arnold HP, Brockmöller J, Johne A, et al. Functional polymorphisms of the human multidrug-resistance gene: multiple sequence variations and correlation of one allele with P-glycoprotein expression and activity in vivo. Proc Natl Acad Sci USA 2000; 97:3473–3478. Leschziner GD, Andrew T, Pirmohamed M, Johnson MR. ABCB1 genotype and PGP expression, function and therapeutic drug response: a critical review and recommendations for future research. Pharmacogenomics J 2007; 7:154–179. Wadelius M, Sörlin K, Wallerman O, Karlsson J, Yue QY, Magnusson PK, et al. Warfarin sensitivity related to CYP2C9, CYP3A5, ABCB1 (MDR1) and other factors. Pharmacogenomics J 2004; 4:40–48. Sussman N, Waltershied M, Butler T, Cali J, Riss T, Kelly J. The predictive nature of high throughput toxicity screening using a human hepatocyte cell line. Cell Notes 2002; 3:7–10.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Acenocoumarol pharmacogenetic algorithm Jiménez-Varo et al. 513

43

44

45

46

47

48 49

50

51

52

Saraeva RB, Paskaleva ID, Doncheva E, Eap CB, Ganev VS. Pharmacogenetics of acenocoumarol: CYP2C9, CYP2C19, CYP1A2, CYP3A4, CYP3A5 and ABCB1 gene polymorphisms and dose requirements. J Clin Pharm Ther 2007; 32:641–649. De Oliveira Almeida VC, De Souza Ferreira AC, Ribeiro DD, Gomes Borges KB, Salles Moura Fernandes AP, Brunialti Godard AL. Association of the C3435T polymorphism of the MDR1 gene and therapeutic doses of warfarin in thrombophilic patients. J Thromb Haemost 2011; 9:2120–2122. Jiménez-Varo E, Cañadas-Garre M, Henriques CI, Pinheiro AM, GutiérrezPimentel MJ, Calleja-Hernández MA. Pharmacogenetics role in the safety of acenocoumarol therapy. Thromb Haemost 2014; 112: [Epub ahead of print]. Rodriguez S, Gaunt TR, Day IN. Hardy–Weinberg equilibrium testing of biological ascertainment for Mendelian randomization studies. Am J Epidemiol 2009; 169:505–514. Schwarz UI, Ritchie MD, Bradford Y, Li C, Dudek SM, Frye-Anderson A, et al. Genetic determinants of response to warfarin during initial anticoagulation. N Engl J Med 2008; 358:999–1008. Levine MN, Raskob G, Landefeld S, Kearon C. Hemorrhagic complications of anticoagulant treatment. Chest 2001; 119 (Suppl):108S–121S. Hylek EM, Go AS, Chang Y, Jensvold NG, Henault LE, Selby JV, Singer DE. Effect of intensity of oral anticoagulation on stroke severity and mortality in atrial fibrillation. N Engl J Med 2003; 349:1019–1026. Thijssen HH, Verkooijen IW, Frank HL. The possession of the CYP2C9*3 allele is associated with low dose requirement of acenocoumarol. Pharmacogenetics 2000; 10:757–760. Choi JR, Kim JO, Kang DR, Yoon SA, Shin JY, Zhang X, et al. Proposal of pharmacogenetics-based warfarin dosing algorithm in Korean patients. J Hum Genet 2011; 56:290–295. Anton AI, Cerezo-Manchado JJ, Padilla J, Perez-Andreu V, Corral J, Vicente V, et al. Novel associations of VKORC1 variants with higher acenocoumarol requirements. PloS One 2013; 8:e64469.

53

54

55

56

57

58

59

60

Kurnik D, Qasim H, Sominsky S, Lubetsky A, Markovits N, Li C, et al. Effect of the VKORC1 D36Y variant on warfarin dose requirement and pharmacogenetic dose prediction. Thromb Haemost 2012; 108:781–788. Loebstein R, Dvoskin I, Halkin H, Vecsler M, Lubetsky A, Rechavi G, et al. A coding VKORC1 Asp36Tyr polymorphism predisposes to warfarin resistance. Blood 2007; 109:2477–2480. Shahin MH, Cavallari LH, Perera MA, Khalifa SI, Misher A, Langaee T, et al. VKORC1 Asp36Tyr geographic distribution and its impact on warfarin dose requirements in Egyptians. Thromb Haemost 2013; 109: 1045–1050. Nahar R, Dube D, Parakh R, Deb R, Saxena R, Singh TP, Verma IC. Implication of novel CYP2C9*57 (p.Asn204His) variant in coumarin hypersensitivity. Thromb Res 2013; 131:535–539. Teichert M, Eijgelsheim M, Rivadeneira F, Uitterlinden AG, van Schaik RH, Hofman A, et al. A genome-wide association study of acenocoumarol maintenance dosage. Hum Mol Genet 2009; 18:3758–3768. Pimenoff VN, Laval G, Comas D, Palo JU, Gut I, Cann H, et al. Similarity in recombination rate and linkage disequilibrium at CYP2C and CYP2D cytochrome P450 gene regions among Europeans indicates signs of selection and no advantage of using tagSNPs in population isolates. Pharmacogenet Genomics 2012; 22:846–857. Carcas AJ, Borobia AM, Velasco M, Abad-Santos F, Díaz MQ, FernándezCapitán C, et al. PGX-ACE Spanish Investigators Group. Efficiency and effectiveness of the use of an acenocoumarol pharmacogenetic dosing algorithm versus usual care in patients with venous thromboembolic disease initiating oral anticoagulation: study protocol for a randomized controlled trial. Trials 2012; 13:239. Verhoef TI, Ragia G, de Boer A, Barallon R, Kolovou G, Kolovou V, et al. EU-PACT Group. A randomized trial of genotype-guided dosing of acenocoumarol and phenprocoumon. N Engl J Med 2013; 369:2304–2312.

Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.