External Evaluation of a Bayesian Warfarin Dose Optimization Based on a Kinetic-Pharmacodynamic Model

Abstract

Warfarin is a representative anticoagulant with large interindividual variability. The published kinetic-pharmacodynamic (K-PD) model allows the prediction of warfarin dose requirement in Swedish patients; however, its applicability in Japanese patients is not known. We evaluated the model’s predictive performance in Japanese patients with various backgrounds and relationships using Bayesian parameter estimation and sampling times. A single-center retrospective observational study was conducted at Tokyo Women’s Medical University, Medical Center East. The study population consisted of adult patients aged >20 years who commenced warfarin with a prothrombin time-international normalized ratio (PT-INR) from June 2015 to June 2019. The published K-PD model modified by Wright and Duffull was assessed using prediction-corrected visual predictive checks, focusing on clinical characteristics, including age, renal function, and individual prediction error. The external dataset included 232 patients who received an initial warfarin daily dose of 3.2 ± 1.28 mg with 2278 PT-INR points (median [range] follow-up period of 23 d [7–28]). Prediction-corrected visual predictive checks carried a propensity for underprediction. Additionally, age >60 years, body mass index ≤25 kg/m², and estimated glomerular filtration rate ≤60 mL/min/1.73 m² had a pronounced tendency to underpredict PT-INR. However, Bayesian prediction using four prior observations reduced underprediction. To improve the prediction performance of these special populations, further studies are required to construct a model to predict warfarin dose requirements in Japanese patients.

INTRODUCTION

Warfarin is a traditional oral anticoagulant that inhibits the inactivation of clotting factor synthesis by vitamin K epoxide reductase in patients with cardiogenic embolism or heart valve replacement.¹⁾ In the clinical setting, warfarin has an extremely large interindividual variability; therefore, clinicians should frequently monitor the international normalized ratio of prothrombin time (PT-INR) as an index of anticoagulation conditions to adjust warfarin dose requirements.²⁾ For instance, the target PT-INR range is 2.0 to 3.0 in patients with stroke.³⁾

The variability in warfarin dose requirement results from age, body size, and genetic variants of CYP2C9 and vitamin K oxide reductase complex (VKORC)1.^4–7) To optimize warfarin dose requirements, pharmacometric models incorporating these variables have been tested in Swedish patients receiving warfarin therapy.^8,9) Bayesian dose optimization using the kinetic-pharmacodynamic (K-PD) model has a tendency to overpredict the maintenance dose at doses of 7 mg/d and higher.¹⁰⁾ However, Japanese patients have a different profile in the genetic variants of CYP2C9 and VKORC1.¹¹⁾ Maintenance doses have been reported to be overestimated at doses above 6 mg/d using the K-PD model in Chinese individuals, who have a similar genetic background to the Japanese.¹²⁾ The predictive performance of the warfarin K-PD model may vary depending on racial and ethnic differences. Additionally, these effects explain only 50% of the individual variability of warfarin.¹³⁾ Recently, profound attention has been focused on the role of renal impairment and drug–drug interactions, such as amiodarone, in the risk of bleeding events associated with warfarin.^14–16) We reported the usefulness of warfarin dose optimization using the K-PD model in Japanese patients by clinical trial simulations,¹⁷⁾ but to the best of our knowledge, there is no report in a real-world dataset that contains patient characteristics such as renal function and concurrent medications.

This study aimed to investigate whether Bayesian dose optimization using the K-PD model was useful for determining the initial warfarin dosage in Japanese patients with various backgrounds, such as those with renal impairment or those concurrently using amiodarone.

MATERIALS AND METHODS

Study Design and Patients

This was a single-center retrospective observational study. The study site was at Tokyo Women’s Medical University, Medical Center East (450 beds, University Hospital, Tokyo, Japan). The study enrolled adult patients aged >20 years who commenced warfarin therapy with PT-INR measurements from June 2015 to June 2019. This study was approved by the institutional review board of Tokyo Women’s Medical University (5307) and Nihon University (19-007) in compliance with the Declaration of Helsinki.

Data Collection

We surveyed the electrical medical chart to generate an external dataset as follows: demographic data (sex, age, height, body weight, fat-free mass,¹⁸⁾ lean body weight,¹⁸⁾ and body mass index), laboratory data (serum albumin, blood urea nitrogen, serum creatinine, and PT-INR), warfarin treatment (dose, dosing interval, and dosing period), and medications of interest (antiplatelet medications and amiodarone). Antiplatelet medications include low-dose aspirin, ticlopidine, clopidogrel, prasugrel, and cilostazol. Renal function was assessed using the prediction equation of estimated glomerular filtration rate (eGFR) as follows: eGFR = 194 × age^−0.287 × serum creatinine^−1.094 × 0.739 (if female).¹⁹⁾ Longitudinal data were extracted from patients who had PT-INR measured at least five times during the first 28 d of treatment. The follow-up period was from the introduction of warfarin therapy to the withdrawal of warfarin therapy or June 2019. Normally distributed data are expressed as mean ± standard deviation, and non-normally distributed data are expressed as median and range. Categorical data were expressed as numbers (%).

K-PD Model of Warfarin

In this study, we applied a published warfarin K-PD model outputting PT-INR from each dose in Swedish⁸⁾ modified by Wright and Duffull²⁰⁾ The previous paper⁸⁾ contains details of this model for S-warfarin, which has three to five times higher anticoagulation potency than R-warfarin. Basic S-warfarin pharmacokinetics describes a 1-compartment model incorporating age and the CYP2C9 genotype. The total clearance of S-warfarin differs according to age and CYP2C9 genotype (*1/*1, *1/*2, *1/*3, *2/*2, *2/*3, or *3/*3). The pharmacodynamic model represents the changes in PT-INR as a function of the S-warfarin dose rate using the E_max model. Two transit compartment chains with three compartments represent the relationship between the S-warfarin dose rate and PT-INR. Additionally, the effective concentration (EC₅₀) differed according to the VKORC1 allele. Wright and Duffull modified this K-PD model to improve PT-INR control without a substantial increase in healthcare resources, such as CYP2C9 and VKORC1 genotypes. The K-PD model is expressed as follows:

  

(1)

  

(2)

  

(3)

  

(4)

  

(5)

  

(6)

  

(7)

  

(8)

  

(9)

  

(10)

  

(11)

  

(12)

where A is the amount of the body; C1_m and C2_m are the mth compartment in the two transit chains, ke is the elimination rate constant, CL and Vd are S-warfarin clearance (0.348 L/h) and distribution volume (14.3 L), DR is the dose rate, EDK₅₀ is the dose rate for 50% inhibition of coagulation, EC₅₀ is the S-warfarin concentration for 50% inhibition of coagulation (4.10 mg/L), EFF is an inhibitory effect, E_max is the maximum inhibition of coagulation set at 1, γ is a Hill coefficient (1.15), MTT₁ and MTT₂ are the mean transit time (28.6 and 118.3 h, respectively), INR_pred is the predicted PT-INR, INR_BASE is the baseline of PT-INR, and INR_max is the theoretical maximum PT-INR set at 20. The inter-subject variability of CL, Vd, and EC₅₀ was 29.8, 23.2, and 33.2 CV%, respectively. The intra-subject variability was 20 CV%. We estimated individual K-PD parameters and simulated PT-INR using “MAXEVAL=0” and the First Order Conditional Estimates Interaction method in the $ESTIMATION step of NONMEM software (version 7.4).

External Evaluation of Kinetic-Pharmacodynamic Model

The external evaluation of the K-PD model performance was used visual and mathematical methods. We visually compared observed against predicted PT-INR by prediction-corrected visual predictive check (pc-VPC)^21,22) with time, demographic data, and laboratory data as independent variables, or stratified based on the presence of concurrent medications, which was generated by simulating 500 replicates of the dataset. The pc-VPC was performed in NONMEM via Wings for NONMEM and the tidyvpc package in R (version 4.0.3).

Next, we assessed the relationship between individual prediction error (IPE) and the number of prior samples using Bayesian prediction to identify the number of prior observations on model performance. The IPE was calculated using the individual predicted values and observed values as the index of prediction capacity. The following is the IPE equation:

  

(13)

where IPE is the individual prediction error, IPRED is the individual predicted PT-INR, and OBS is the observed PT-INR. The percentage of absolute IPEs within 20% was assessed for each sampling time. The number of samples using Bayesian estimation was set from zero to five.

RESULTS

Patient Backgrounds

This study included 232 patients who underwent warfarin therapy. The patient characteristics are summarized in Table 1. The study patients had a median age (range) of 72 (26–96) years. The median eGFR was 47.4 (2.6–199.0) mL/min/1.73 m². There were 178 (76.7%) patients who met the criteria for chronic kidney disease. Forty-five patients (19.4%) received concurrent antiplatelet therapy. Twenty-three patients (9.9%) received concurrent amiodarone therapy. The initial daily dose of warfarin was 3.2 ± 1.28 mg with a median follow-up period of 23 d (7–28). A total of 2278 PT-INR measurements were available for analysis.

Table 1. Patient Characteristics

Characteristics	N = 232
Demographical
Male, N (%)	131 (56.5)
Age, years	72 [26–96]
Age >60 years, N (%)	185 (79.7)
Height, cm	160 [133–185]
Body weight, kg	54.6 [33.0–103.0]
Lean body weighta), kg	42.2 [24.6–72.3]
Fat free massa), kg	44.9 [25.8–69.4]
BMI, kg/m2	21.7 [14.0–36.0]
BMI ≤25 kg/m2, N (%)	185 (79.7)
Laboratory data
Serum albumin, g/dL	2.9 [1.4–4.7]
Blood urea nitrogen, mg/dL	18.6 [4.9–98.9]
Serum creatinine, mg/dL	0.94 [0.31–13.2]
eGFRb), mL/min/1.73 m2	47.4 [2.6–199.0]
eGFRb) ≤60 mL/min/1.73 m2, N (%)	178 (76.7)
Warfarin
Initial warfarin daily dose, mg	3 [0.5–10]
Medications of interest
Antiplatelet medicationsc), N (%)	45 (19.4)
Low-dose aspirin, N (%)	34 (14.7)
Ticlopidine, N (%)	1 (0.4)
Clopidogrel, N (%)	7 (3.0)
Prasugrel, N (%)	0 (0.0)
Cilostazol, N (%)	8 (3.4)
Amiodarone, N (%)	23 (9.9)

Data represents median [range] or number (%). Abbreviations: N, number; BMI, body mass index; eGFR, estimated glomerular filtration rate. a) Fat-free mass (kg) was calculated using the following equations: 0.285 × BW (kg) + 12.1 × HT (m²) for males 0.287 × BW (kg) + 9.74 × HT (m)² for females. Lean body weight (kg) was calculated using the following equations: 1.1 × BW (kg)−0.0128 × BMI (kg/m²) × BW (kg) for males. 1.07 × BW (kg)−0.0148 × BMI (kg/m²) × BW (kg) for females where BW, BMI, and HT represent body weight, body mass index, and height, respectively.¹⁸⁾ b) The prediction formula of eGFR was as follows: eGFR = 194 × age^−0.287 × serum creatinine^−1.094 × 0.739 (if female).¹⁹⁾ c) Antiplatelet medications include low-dose aspirin, ticlopidine, clopidogrel, prasugrel, and cilostazol.

K-PD Model Evaluation

A prediction-corrected visual predictive check is shown in Fig. 1. Time courses of predicted PT-INR values were lower than those of observed PT-INR values across all measurement points. Additionally, Fig. 2 shows pc-VPC with age, body mass index, serum albumin, urea nitrogen, serum creatinine, and eGFR as independent variables. Patients aged >60 years, with body mass index ≤25 kg/m², and eGFR ≤60 mL/min/1.73 m² have a tendency toward lower predicted PT-INR. The other pc-VPCs are shown in Supplementary Fig. 1S. Additionally, the underprediction of PT-INR was consistent with a lower body surface area-unstandardized eGFR (mL/min), lower lean body weight, and lower fat-free mass, as shown in Supplementary Fig. 1S. Figure 3 shows pc-VPC stratified by the presence of concomitant drugs. Concomitant antiplatelet drugs have a tendency toward a lower predicted PT-INR.

Fig. 1. Prediction-Corrected Visual Predictive Check

X-axis and Y-axis represent the warfarin therapy period and prediction-corrected observed and predicted values, respectively. Points represent observed values.

Fig. 2. Prediction-Corrected Visual Predictive Check with Age, Serum Albumin, Body Mass Index, eGFR, Blood Urea Nitrogen, and Serum Creatinine as Independent Variables

Fig. 3. Prediction-Corrected Visual Predictive Check Stratified on the Presence of Concomitant Use of Drugs

X-axis and Y-axis represent the warfarin therapy period and prediction-corrected observed and predicted values, respectively. Upper panel, stratified on the presence of concomitant use of amiodarone; lower panel, stratified on the presence of concomitant use of antiplatelet drugs. Antiplatelet drugs include low-dose aspirin, ticlopidine, clopidogrel, prasugrel, and cilostazol.

The relationships between IPE and sampling times are shown in Fig. 4. The results of the four prior observations are nearly identical to those of all prior observations. The percentages within 20% of IPE% for 0–5 sampling times were 46, 47, 51, 56, 58, and 60%, respectively. IPE was biased in a negative direction for sampling from zero to three times. We ascertained that the improvement in the percentages within 20% of IPE% was accompanied by a higher number of Bayesian predictions. Stratification by antiplatelet use, patient age ≤60 years, body mass index >25 kg/m², and eGFR >60 mL/min/1.73 m² showed a decreasing bias in IPE, with the impact being more pronounced for age and body mass index.

Fig. 4. Summary of Box Plots of Individual Prediction Error with Bayesian Estimation for PT-INR

A, all patients; B–E; stratified on the presence of concomitant use of antiplatelet therapy, eGFR (≤60 mL/min/1.73 m² or >60 mL/min/1.73 m²), age (≤60 years or >60 years), and BMI (≤25 kg/m² or >25 kg/m²). The median is indicated by a horizontal line within the box. The bottom and top edges of the box represent the 25th and 75th percentiles, respectively. The maximum length of each whisker is 1.5 times the interquartile range. Outlier is not shown. The dashed lines represent ±20%.

DISCUSSION

This study revealed that the K-PD model for warfarin in Swedish patients could underpredict PT-INR in Japanese patients. This underprediction was notable in patients with renal impairment and low body mass index, elderly patients, and patients using antiplatelet therapy. The prediction accuracy improved with a large number of measurement points.

The applied K-PD model included significant relevant covariates, including age and genetic variants of CYP2C9 and VKORC1.⁸⁾ Generally, age has a negative correlation with S-warfarin clearance with large variability, which makes it difficult to interpret our findings.⁶⁾ Age might have a strong impact on the elevation of PT-INR during warfarin treatment beyond expectations. We consider that genetic variants were the predominant reason for the dissociation between the predicted and observed PT-INR values. In addition, the influence of low body mass index on prediction accuracy was notable in the present study. Indeed, a previous study demonstrated that a small body size was associated with an increased risk of bleeding associated with warfarin therapy.^14,23,24) Hence, the present study suggests that a low body mass index is an important factor in tailoring warfarin dose requirements. These findings highlight the need to construct a K-PD model that is suitable for Japanese patients who undergo warfarin therapy.

The kidneys play an important role in drug elimination. However, previous research has demonstrated that free clearance decreases with advancing chronic kidney disease stage in CYP2C9 substrates.²⁵⁾ Furthermore, warfarin dose requirement was correlated with creatinine clearance, adjusting for the VKORC1 genotype.²⁶⁾ These reports strongly support that renal function affects not only pharmacokinetics but also pharmacodynamics. Therefore, it is difficult to determine the appropriate warfarin dose requirement for patients with renal impairment. The modified K-PD model required further refinement to optimize warfarin dose requirement in patients with renal impairment, in addition to elderly patients and patients with low body mass index. Older age, lower renal function, lower body mass index, lower lean body weight, and lower fat-free mass, which were found to reduce the predictive performance in this study, are characteristics consistent with those of patients with sarcopenia. In oral anticoagulants, lean body weight was reported as a variable in direct oral anticoagulant levels in patients with sarcopenia.²⁷⁾ Dose–response analyses using warfarin dose and PT-INR have also reported an increased responsiveness with a decreased fat-free mass.²⁸⁾ The creatinine used to calculate eGFR also varies with muscle mass. Muscle strength measures, such as grip strength and short physical performance battery, and renal function measures, such as cystatin C, which are not affected by muscle mass, should be considered as potential covariates in a K-PD model for Japanese patients.

Warfarin interacts with amiodarone and N-diethyl amiodarone, the active metabolite of amiodarone, resulting in the elevation of S-warfarin concentration by inhibiting CYP2C9.^29,30) This effect remains long-term because amiodarone and N-diethyl amiodarone have a long half-life. However, we could not detect it because only 9.9% of all cases were treated with amiodarone. In addition, antiplatelet medications slightly affect the prediction accuracy. Co-administration of antiplatelet medication increases the risk of bleeding due to pharmacodynamic alteration.¹⁴⁾ In the present study, as shown in Fig. 3, there was a trend towards the underprediction of PT-INR with the concomitant use of antiplatelet agents. However, clopidogrel, one of the antiplatelet agents studied in this study, has not been found to interact with warfarin.³¹⁾ The impact of each antiplatelet agent on the predictive performance needs to be investigated in the future. More than 30 algorithms to predict warfarin dose requirement did not include amiodarone as a clinical covariate, which can predict warfarin dose requirements of approximately 50%.²⁾ Thus, we believe that concurrent medications are a clinically significant factor for warfarin dose requirements.

Several reports support that Bayesian dose optimization is useful for improving the prediction capacity of the population pharmacokinetic model.²⁰⁾ As the number of PT-INR measurements increased, we confirmed that the Bayesian prediction produced a better prediction capacity. Bayesian prediction using four PT-INR measurements showed that 58% of cases were within IPE ±20%, and underprediction was reduced. This result is comparable to that of the prediction using all the measurements. The K-PD model tended to underpredict PT-INR in Japanese patients. However, as shown in Fig. 4, the percentage of patients with IPE ±20% increased by Bayesian estimation, taking into account the presence of concomitant antiplatelet agents, eGFR, age, and body mass index. Especially for patients aged ≤60 years with a body mass index >25 kg/m², Bayesian estimation of two PT-INR measurements showed that about half of the patients had an IPE of ±20%. For a more accurate prediction, it would be desirable to build a K-PD model for the Japanese population, but we believe that the K-PD model reported in this study has clinical applicability. Thus, this method is valuable in the early stages of warfarin therapy introduction. Although the measurement interval of PT-INR was inconsistent, further studies are warranted to establish the best timing.

Predictions from the warfarin K-PD model in Chinese patients have been reported to overpredict maintenance doses at dosing rates above 6 mg/d.¹²⁾ Because this study included patients in the induction phase of warfarin therapy, data at a daily dose of 6 mg/d or higher were observed in 4 patients (1.7%), and the relationships between the predictability and high dose could not be observed in the present study. A study of ethnic differences in the pharmacokinetics of meloxicam, which is metabolized by CYP2C9 in the same manner as warfarin, did not detect any ethnic differences in pharmacokinetics between Japanese and Chinese patients.²¹⁾ Therefore, Japanese, as well as Chinese, may overpredict the maintenance dose at dosing rates above 6 mg/d in the maintenance stage. The investigation of maintenance doses in the Japanese population is a subject for future investigation.

This study had several limitations in interpreting the results. First, this was a single-center, retrospective, observational study that did not pursue all potential confounding factors. Second, we could not quantify medication adherence based on the study design. Additionally, other clinical backgrounds (e.g., actual dietary intake and cirrhosis) were not evaluated. Third, the genetic variants of CYP2C9 and VKORC1 are unknown. Therefore, we decided not to perform further analyses because we could not modify the K-PD model. Finally, the plans for warfarin dose adjustment varied.

The K-PD model for warfarin in Swedish patients underpredicts PT-INR in Japanese patients with renal impairment and low body mass index, elderly patients, and patients using antiplatelet therapy, which highlights the need for a renal-function-guided K-PD model. Further studies are required to construct a model to predict warfarin dose requirements in Japanese patients.

Acknowledgment

This work was supported by JSPS KAKENHI (Grant No. JP20K16092).

Conflict of Interest

The authors declare no conflict of interest.

Supplementary Materials

The online version of this article contains supplementary materials.

REFERENCES

• 1) Hirsh J, Dalen J, Anderson DR, Poller L, Bussey H, Ansell J, Deykin D. Oral anticoagulants: mechanism of action, clinical effectiveness, and optimal therapeutic range. Chest, 119 (Suppl.), 8S–21S (2001).
• 2) Verhoef TI, Redekop WK, Daly AK, van Schie RM, de Boer A, Maitland-van der Zee AH. Pharmacogenetic-guided dosing of coumarin anticoagulants: algorithms for warfarin, acenocoumarol and phenprocoumon. Br. J. Clin. Pharmacol., 77, 626–641 (2014).
• 3) Reynolds MW, Fahrbach K, Hauch O, Wygant G, Estok R, Cella C, Nalysnyk L. Warfarin anticoagulation and outcomes in patients with atrial fibrillation: a systematic review and metaanalysis. Chest, 126, 1938–1945 (2004).
• 4) Wattanachai N, Kaewmoongkun S, Pussadhamma B, Makarawate P, Wongvipaporn C, Kiatchoosakun S, Vannaprasaht S, Tassaneeyakul W. The impact of non-genetic and genetic factors on a stable warfarin dose in Thai patients. Eur. J. Clin. Pharmacol., 73, 973–980 (2017).
• 5) Sconce EA, Khan TI, Wynne HA, Avery P, Monkhouse L, King BP, Wood P, Kesteven P, Daly AK, Kamali F. The impact of CYP2C9 and VKORC1 genetic polymorphism and patient characteristics upon warfarin dose requirements: proposal for a new dosing regimen. Blood, 106, 2329–2333 (2005).
• 6) Jensen BP, Chin PK, Roberts RL, Begg EJ. Influence of adult age on the total and free clearance and protein binding of (R)- and (S)-warfarin. Br. J. Clin. Pharmacol., 74, 797–805 (2012).
• 7) Jorgensen AL, Al-Zubiedi S, Zhang JE, Keniry A, Hanson A, Hughes DA, Eker D, Stevens L, Hawkins K, Toh CH, Kamali F, Daly AK, Fitzmaurice D, Coffey A, Williamson PR, Park BK, Deloukas P, Pirmohamed M. Genetic and environmental factors determining clinical outcomes and cost of warfarin therapy: a prospective study. Pharmacogenet. Genomics, 19, 800–812 (2009).
• 8) Hamberg AK, Wadelius M, Lindh JD, Dahl ML, Padrini R, Deloukas P, Rane A, Jonsson EN. A pharmacometric model describing the relationship between warfarin dose and INR response with respect to variations in CYP2C9, VKORC1, and age. Clin. Pharmacol. Ther., 87, 727–734 (2010).
• 9) Hamberg AK, Dahl ML, Barban M, Scordo MG, Wadelius M, Pengo V, Padrini R, Jonsson EN. A PK-PD model for predicting the impact of age, CYP2C9, and VKORC1 genotype on individualization of warfarin therapy. Clin. Pharmacol. Ther., 81, 529–538 (2007).
• 10) Saffian SM, Duffull SB, Roberts RL, Tait RC, Black L, Lund KA, Thomson AH, Wright DF. Influence of genotype on warfarin maintenance dose predictions produced using a bayesian dose individualization tool. Ther. Drug Monit., 38, 677–683 (2016).
• 11) Takahashi H, Wilkinson GR, Nutescu EA, Morita T, Ritchie MD, Scordo MG, Pengo V, Barban M, Padrini R, Ieiri I, Otsubo K, Kashima T, Kimura S, Kijima S, Echizen H. Different contributions of polymorphisms in VKORC1 and CYP2C9 to intra- and inter-population differences in maintenance dose of warfarin in Japanese, Caucasians and African-Americans. Pharmacogenet. Genomics, 16, 101–110 (2006).
• 12) Dong J, Shi GH, Lu M, Huang S, Liu YH, Yao JC, Li WY, Li LX. Evaluation of the predictive performance of Bayesian dosing for warfarin in Chinese patients. Pharmacogenomics, 20, 167–177 (2019).
• 13) Wadelius M, Chen LY, Lindh JD, Eriksson N, Ghori MJ, Bumpstead S, Holm L, McGinnis R, Rane A, Deloukas P. The largest prospective warfarin-treated cohort supports genetic forecasting. Blood, 113, 784–792 (2009).
• 14) Hirai T, Hamada Y, Geka Y, Kuwana S, Hirai K, Ishibashi M, Fukaya Y, Kimura T. A retrospective study on the risk factors for bleeding events in warfarin therapy, focusing on renal function. Eur. J. Clin. Pharmacol., 73, 1491–1497 (2017).
• 15) Olesen JB, Lip GY, Kamper AL, Hommel K, Køber L, Lane DA, Lindhardsen J, Gislason GH, Torp-Pedersen C. Stroke and bleeding in atrial fibrillation with chronic kidney disease. N. Engl. J. Med., 367, 625–635 (2012).
• 16) Jun M, James MT, Manns BJ, Quinn RR, Ravani P, Tonelli M, Perkovic V, Winkelmayer WC, Ma Z, Hemmelgarn BR. The association between kidney function and major bleeding in older adults with atrial fibrillation starting warfarin treatment: population based observational study. BMJ, 350 (feb03 4), h246 (2015).
• 17) Aoyama T, Nojiri K, Takada K, Kimura R, Miyamoto A, Kose E, Hayashi H, Matsumoto Y. Comparison of nomograms and Bayesian dose individualization for warfarin therapy in virtual Japanese patients. Jpn. J. TDM, 33, 15–33 (2016).
• 18) Green B, Duffull SB. What is the best size descriptor to use for pharmacokinetic studies in the obese? Br. J. Clin. Pharmacol., 58, 119–133 (2004).
• 19) Matsuo S, Imai E, Horio M, Yasuda Y, Tomita K, Nitta K, Yamagata K, Tomino Y, Yokoyama H, Hishida A. Revised equations for estimated GFR from serum creatinine in Japan. Am. J. Kidney Dis., 53, 982–992 (2009).
• 20) Wright DF, Duffull SB. A Bayesian dose-individualization method for warfarin. Clin. Pharmacokinet., 52, 59–68 (2013).
• 21) Aoyama T, Ishida Y, Kaneko M, Miyamoto A, Saito Y, Tohkin M, Kawai S, Matsumoto Y. Pharmacokinetics and pharmacodynamics of meloxicam in East Asian populations: the role of ethnicity on drug response. CPT Pharmacometrics Syst. Pharmacol., 6, 823–832 (2017).
• 22) Jamsen KM, Patel K, Nieforth K, Kirkpatrick CMJ. A regression approach to visual predictive checks for population pharmacometric models. CPT Pharmacometrics Syst. Pharmacol., 7, 678–686 (2018).
• 23) Hart R, Veenstra DL, Boudreau DM, Roth JA. Impact of body mass index and genetics on warfarin major bleeding outcomes in a community setting. Am. J. Med., 130, 222–228 (2017).
• 24) Sandhu RK, Ezekowitz J, Andersson U, Alexander JH, Granger CB, Halvorsen S, Hanna M, Hijazi Z, Jansky P, Lopes RD, Wallentin L. The ‘obesity paradox’ in atrial fibrillation: observations from the ARISTOTLE (Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation) trial. Eur. Heart J., 37, 2869–2878 (2016).
• 25) Tan ML, Yoshida K, Zhao P, Zhang L, Nolin TD, Piquette-Miller M, Galetin A, Huang SM. Effect of chronic kidney disease on nonrenal elimination pathways: a systematic assessment of CYP1A2, CYP2C8, CYP2C9, CYP2C19, and OATP. Clin. Pharmacol. Ther., 103, 854–867 (2018).
• 26) Ichihara N, Ishigami T, Umemura S. Effect of impaired renal function on the maintenance dose of warfarin in Japanese patients. J. Cardiol., 65, 178–184 (2015).
• 27) Bendayan M, Mardigyan V, Williamson D, Chen-Tournoux A, Eintracht S, Rudski L, MacNamara E, Blostein M, Afilalo M, Afilalo J. Muscle mass and direct oral anticoagulant activity in older adults with atrial fibrillation. J. Am. Geriatr. Soc., 69, 1012–1018 (2021).
• 28) Xue L, Zhang Y, Xie C, Zhou L, Liu L, Zhang H, Xu L, Song H, Lin M, Qiu H, Zhu J, Zhu Y, Zou J, Zhuang W, Xuan B, Chen Y, Fan Y, Wu D, Shen Z, Miao L. Relationship between warfarin dosage and international normalized ratio: a dose-response analysis and evaluation based on multicenter data. Eur. J. Clin. Pharmacol., 75, 785–794 (2019).
• 29) Sanoski CA, Bauman JL. Clinical observations with the amiodarone/warfarin interaction: dosing relationships with long-term therapy. Chest, 121, 19–23 (2002).
• 30) McDonald MG, Au NT, Wittkowsky AK, Rettie AE. Warfarin-amiodarone drug–drug interactions: determination of [I](u)/K(I,u) for amiodarone and its plasma metabolites. Clin. Pharmacol. Ther., 91, 709–717 (2012).
• 31) Lidell C, Svedberg LE, Lindell P, Bandh S, Job B, Wallentin L. Clopidogrel and warfarin: absence of interaction in patients receiving long-term anticoagulant therapy for non-valvular atrial fibrillation. Thromb. Haemost., 89, 842–846 (2003).