Edoxaban Population Pharmacokinetics in Chinese Patients with Nonvalvular Atrial Fibrillation: Model-Informed Dose Adjustment

Abstract

Background

Edoxaban is a novel oral anticoagulant that directly inhibits factor Xa. The ENGAGE population pharmacokinetic (PopPK) model was established using data from the ENGAGE AF-TIMI 48 trial, which included over 21,000 participants, predominantly (81%) White/Caucasian individuals. However, its applicability to Chinese patients necessitates further evaluation. This study assessed the suitability of the ENGAGE PopPK model for Chinese patients with nonvalvular atrial fibrillation (NVAF).

Methods

We analyzed 730 pharmacokinetic (PK) plasma samples from 104 Chinese NVAF patients using nonlinear mixed-effects modeling.

Results

A two-compartment model with first-order absorption and linear elimination optimally described the PK data of edoxaban in this cohort. Body weight (WT) and creatinine clearance were identified as significant covariates of apparent clearance (CL/F), and positively correlated with CL/F. Model-based simulations demonstrated that, in the 30 mg once-daily (q.d.) dosing group, patients with moderate renal impairment and low WT experienced higher systemic exposure than those with only one of these attributes. Additionally, patients with moderate renal impairment, whether alone or combined with low WT, displayed elevated steady-state trough concentrations (C_min,ss) compared to those receiving the 60 mg q.d. dose.

Conclusion

Our PopPK model characterizes the PK profile of edoxaban in Chinese NVAF patients, identifies critical covariates influencing drug exposure, and proposes an evidence-based dosing regimen tailored to this population. This trial was registered at www.clinicaltrials.gov as #NCT05320627.

Keywords

edoxaban - PopPK model - nonvalvular atrial fibrillation - dosing regimen - bleeding risk

Introduction

Edoxaban, a novel oral anticoagulant, selectively targets factor Xa (FXa) to prevent thrombus formation by inhibiting free FXa and diminishing thrombin generation. Additionally, it suppresses prothrombinase activity, thereby mitigating thrombin-mediated platelet aggregation. Edoxaban is a preferred therapeutic option for stroke prevention in patients with nonvalvular atrial fibrillation (NVAF).[1] [2] [3]

Following oral administration, edoxaban is rapidly absorbed, achieving peak plasma concentrations within 1 to 2 hours.[4] Although food intake may variably elevate peak concentrations, its impact on total systemic exposure remains negligible.[5] The absolute bioavailability of edoxaban is approximately 62%. Edoxaban is characterized by a biphasic distribution pattern. Renal excretion serves as the primary elimination pathway, contributing a total body clearance of approximately 11 L/hour. Renal clearance accounts for about 35% of the administered dose, with the remainder cleared via metabolism and biliary/fecal routes. Edoxaban serves as a substrate for the efflux transporter P-glycoprotein (P-gp).[6] [7] Clinical studies have shown that coadministration with P-gp inhibitors, such as quinidine and cyclosporine, augments edoxaban exposure.[8]

The ENGAGE AF-TIMI 48 study, a large-scale, multinational, multicenter phase III trial, compared the efficacy and safety of edoxaban versus warfarin in NVAF patients. A population pharmacokinetic (PopPK) model was developed using data from multiple phase I and II studies and the ENGAGE AF-TIMI 48 trial.[9] [10] [11] [12] [13] This model comprehensively characterized the pharmacokinetic (PK) variability of edoxaban in NVAF patients, identifying key covariates significantly influencing its PK. Notably, low body weight (WT), moderate renal impairment, and concomitant P-gp inhibitors markedly increased edoxaban exposure, prompting a 50% dose reduction in affected patients to mitigate associated risks.[14] [15] [16] The ENGAGE PopPK model was further investigated by Shimizu et al, who incorporated data from 90 subjects with severe renal impairment (SRI) into the existing ENGAGE study data.[17] The subsequent research demonstrated that the ENGAGE PopPK model is applicable to NVAF patients with SRI.

Additionally, a PopPK model derived from the Hokusai VTE study indicated a pronounced increase in clearance/bioavailability and central volume of distribution among Asian patients,[18] [19] suggesting potential ethnic variations in the covariates governing the PK profile of edoxaban. Nonetheless, the PopPK properties of edoxaban in Chinese NVAF patients remain inadequately elucidated. It is uncertain whether existing PopPK models for edoxaban can accurately characterize the PK profile of this population and inform optimal dosing regimens.

To address this gap, we conducted a prospective clinical study to evaluate the applicability of the ENGAGE PopPK model in Chinese NVAF patients. The primary aim was to determine whether this model adequately describes the PK behavior of edoxaban in this cohort. Should it prove insufficient, the study sought to construct a novel PopPK model tailored to Chinese NVAF patients. This new model aims to identify critical covariates driving variability in edoxaban exposure and to give an evidence-based dosing regimen for potential clinical application in this population.

Methods

Clinical Studies

The PK data for this study were derived from a prospective, multicenter, open-label clinical trial of edoxaban, registered at ClinicalTrials.gov (identifier: NCT05320627). The trial enrolled male and female patients with NVAF, aged 20 years or older, who met the clinical indications for edoxaban therapy. Participants were assigned to one of three edoxaban (LIXIANA tablets, Daiichi Sankyo, Inc., Nihonbashi-Honcho, Chuo-Ku, Tokyo, Japan) dosing regimens: 15, 30, or 60 mg, based on the following criteria:

1. A total of 60 mg once daily (Quaque die, q.d.): patients with normal or mild renal impairment (creatinine clearance [CrCl] ≥ 50 mL/minute).

2. A total of 30 mg q.d.: patients exhibiting one or more of the following: moderate or SRI (15 mL/minute < CrCl < 50 mL/minute), WT ≤ 60 kg, or concurrent use of P-gp inhibitors.

3. A total of 15 mg q.d.: patients aged 80 years or older with one or more risk factors, including at least one high bleeding risk factor, low WT (≤45 kg), SRI (15 mL/minute < CrCl < 30 mL/minute), chronic use of nonsteroidal anti-inflammatory drugs, or use of antiplatelet agents.

Exclusion criteria included a history of mechanical valve replacement, atrial fibrillation unrelated to NVAF, moderate or severe anemia, CrCl < 15 mL/minute or dialysis dependence, use of antiplatelet medications, contraindications to edoxaban, life expectancy below 6 months, uncontrolled hypertension, hepatic impairment, elevated bleeding risk, positive urine pregnancy test in women of childbearing potential, or participation in another drug or medical device trial within 30 days prior to screening.

The treatment duration was up to 12 weeks. Blood samples for PK analysis were collected at baseline, predose, and between 1 and 8 hours postdose during visits at weeks 4, 8, or 12. Safety assessments were conducted at each follow-up visit.

Covariates deemed clinically relevant included sex, age, WT, body mass index (BMI), body surface area (BSA), serum creatinine (CREAT), alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBIL), concomitant use of P-gp inhibitors, and CrCl calculated via the Cockcroft–Gault equation. Missing continuous covariate values were imputed with the study population median, while missing categorical covariates were assigned the most prevalent category or designated as “missing.”

All participants provided written informed consent prior to enrollment. The trial protocol and amendments received approval from the Institutional Review Board and Independent Ethics Committee of Fuwai Hospital, Chinese Academy of Medical Sciences. The study adhered to the principles of the Declaration of Helsinkit and all applicable amendments, as well as the International Conference on Harmonization Guidelines for Good Clinical Practice (GCP).

Bioassays

In this study, edoxaban concentrations in plasma were quantified using a validated high-performance liquid chromatography-tandem mass spectrometry method. Linear calibration curves spanned 1 to 500 ng/mL, with an LLOQ of 1 ng/mL. Intra-assay precision was 13%, and inter-assay precision was 12%. Intra-assay accuracy ranged from −4 to 7%, while inter-assay accuracy ranged from 0 to 5%.

PopPK Modeling

Software

The PopPK model was developed using NONMEM 7.5.0 (ICON Development Solutions, Ellicott City, Maryland, United States) with the gFortran compiler (version 4.6.3) and PsN 5.2.6 (Department of Pharmaceutical Biosciences, Uppsala University, Sweden). R 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) facilitated raw data organization, NONMEM dataset creation, exploratory data analysis, and model evaluation with visual outputs.

External Validation of the ENGAGE Model

External validation of the ENGAGE PopPK model was performed using empirical Bayes estimation of individual PK parameters, incorporating new PK data from Chinese NVAF patients. Model performance was assessed via standard diagnostic tools, including prediction error, goodness-of-fit (GOF) plots, and prediction corrected visual predictive check (pc-VPC).

Base Model

The PopPK model was constructed using a nonlinear mixed effects model for edoxaban plasma concentration-time data. Parameter estimation was performed using the first-order conditional estimation with inter- and intra-subject variability interaction. Initial fits tested one- and two-compartment models with first-order absorption, linear elimination, and a fixed WT effect on PK parameters to establish the structural framework. Subsequent refinements explored absorption dynamics, including lag time or zero-order absorption. Apparent clearance (CL/F), comprising renal and nonrenal clearance, as demonstrated in the ENGAGE model, was tried in the base model as well. The influence of P-gp inhibitors on bioavailability (F) was also examined. Given the known processes of absorption, distribution, metabolism, and elimination of edoxaban, a sparse sampling approach and estimation outcomes, a flip-flop kinetic model was considered, wherein the absorption rate constant (Ka) is governed by the elimination rate constant (Kel), as expressed in equations [(1)] and [(2)]:

Model selection was based on the objective function value (OFV), parameter estimate precision, and GOF plots.

Inter-individual variability in PK parameters was modeled using an exponential function (equation [[3]]).

Where P _i represents the individual parameter for subject i, represents the population parameter. η _i is the value of inter-individual variability for the ith individual based on empirical Bayes estimates.

Residual unexplained variability (RUV) was assessed using proportional (equation [[4]]), additive (equation [[5]]), or combined (equation [[6]]) models:

Where Y _ij is the observation, F _ij is the individual predicted concentration (IPRED), and ε₁, _ij represents the proportional error components. ε_{2, ij} represents the additive error components.

Covariate Model

Covariate effects were prescreened by correlating post hoc Bayesian estimates with covariates using linear regression for continuous variables and analysis of variance for categorical variables, with statistical significance set at p <0.01. Significant covariates underwent stepwise forward inclusion and backward deletion. In forward inclusion, a covariate was retained if it reduced the OFV by > 6.63 (χ ² test, degree of freedom [df] = 1, p < 0.01). In backward deletion, covariates increasing the OFV by > 10.83 (χ ² test, df = 1, p < 0.001) were retained. Continuous covariates were modeled with a power function (equation [[7]]):

Categorical covariates used a proportional function (equation [[8]]):

Where P _TV is the population typical value of the parameter, P _i is the subpopulation parameter, COV is the continuous covariate, COV _med is the median of the continuous covariate, and θ is the coefficient of the effect of the covariate on the parameter.

Model Evaluation

The final model was evaluated using GOF plots and pc-VPC. A pc-VPC with 1,000 simulations calculated the median, 2.5th, and 97.5th percentiles of simulated concentrations, overlaid with observed data for assessment.

Simulation of Exposure Measures in Chinese NVAF Patients

Monte Carlo simulations, based on the final model, predicted steady-state exposure metrics, including maximum concentration (C_max,ss), trough concentration (C_min,ss), area under the plasma concentration–time curve (AUC_ss), average plasma concentration (C_avg,ss), and concentration–time profile. Deterministic simulations elucidated exposure variations across virtual patient cohorts.[20] Based on prescribing guidelines and model outcomes, three key factors (WT, CrCl, and age) defined eight simulated patient groups: standard WT (70 kg) and low WT (50 kg)[9] [21] [22] [23]; normal renal function (CrCl = 84.35 mL/minute), moderate renal impairment (CrCl = 35 mL/minute), and SRI (CrCl = 20 mL/minute)[17] [24] [25] [26] [27]; and standard age (65 years) and advanced age (80 years).[16] [28]

Among these, group 1 received 60 mg q.d. (standard: WT 70 kg, CrCl 84.35 mL/minute, age 65 years), while group 2 received 30 mg q.d. (same profile). Groups 3 to 5 received 30 mg q.d. with dose reductions for: group 3 (moderate renal impairment, CrCl = 35 mL/minute), group 4 (low WT, 50 kg), and group 5 (combined moderate renal impairment and low WT). Groups 6 to 8 received 15 mg q.d., incorporating advanced age (80 years) with: group 6 (SRI, CrCl = 20 mL/minute), group 7 (low WT, 50 kg), and group 8 (combined SRI and low WT).

Results

Demographics

A total of 730 plasma samples for PK analysis were obtained from 104 Chinese patients with NVAF in this study. The cohort was predominantly male (61%), with most participants (92%) not receiving concomitant P-gp inhibitors. The median age was 65 years (range: 36–85), median WT was 69.5 kg (range: 45–108), and median CrCl was 84.35 mL/minute (range: 16.70–183.0). The demographics of the patients are detailed in [Table 1].

Table 1 The demographics and biological baseline characteristics of Chinese NVAF patients with edoxaban

Characteristics of the participants	15 mg (n = 4)	30 mg (n = 31)	60 mg (n = 69)	Overall (n = 104)
Sex, n (%)
Female	2 (50%)	22 (71%)	17 (25%)	41 (39%)
Male	2 (50%)	9 (29%)	52 (75%)	63 (61%)
Age, y
Mean (SD)	82.5 (2.38)	62.9 (9.33)	61.7 (9.72)	62.8 (10.2)
Median [min, max]	82.5 [80.0, 85.0]	61.0 [46.0, 80.0]	65.0 [36.0, 78.0]	65.0 [36.0, 85.0]
WT, kg
Mean (SD)	62.3 (5.56)	58.0 (8.87)	75.4 (10.1)	69.7 (12.5)
Median [min, max]	61.0 [57.0, 70.0]	58.0 [45.0, 83.0]	75.0 [60.0, 108]	69.5 [45.0, 108]
BMI, kg/m2
Mean (SD)	25.2 (2.12)	22.3 (2.90)	25.8 (2.85)	24.7 (3.26)
Median [min, max]	25.2 [23.7, 26.7]	21.8 [15.4, 28.7]	25.5 [20.5, 35.3]	24.8 [15.4, 35.3]
Missing	2 (50%)	0 (0%)	0 (0%)	2 (2%)
BSA, m2
Mean (SD)	1.55 (0)	1.60 (0.135)	1.87 (0.139)	1.78 (0.186)
Median [min, max]	1.55 [1.55, 1.55]	1.59 [1.39, 1.95]	1.86 [1.59, 2.22]	1.81 [1.39, 2.22]
ALT, IU/L
Mean (SD)	11.0 (4.24)	34.2 (24.4)	27.5 (21.8)	29.2 (22.6)
Median [min, max]	11.0 [8.00, 14.0]	26.0 [7.00, 113]	24.0 [1.00, 173]	24.0 [1.00, 173]
Missing	2 (50%)	0 (0%)	0 (0%)	2 (2%)
AST, IU/L
Mean (SD)	16.4 (2.42)	31.9 (11.2)	27.8 (12.6)	28.7 (12.3)
Median [min, max]	16.0 [14.2, 19.0]	30.0 [13.0, 54.0]	25.0 [11.0, 109]	27.0 [11.0, 109]
Missing	1 (25%)	0 (0%)	0 (0%)	1 (1%)
CrCl, mL/min
Mean (SD)	33.20 (24.0)	79.37 (27.5)	98.93 (31.2)	90.57 (33.0)
Median [min, max]	23.60 [16.70, 68.90]	75.20 [35.00, 161.0]	92.10 [51.00, 183.0]	84.35 [16.70, 183.0]
CREAT, μmol/L
Mean (SD)	165 (77.7)	67.1 (26.1)	74.0 (14.5)	75.4 (29.0)
Median [min, max]	193 [51.0, 222]	62.3 [40.6, 191]	72.1 [40.4, 122]	69.2 [40.4, 222]
TBIL, μmol/L
Mean (SD)	6.63 (1.05)	12.1 (4.74)	13.9 (6.02)	13.2 (5.71)
Median [min, max]	6.63 [5.89, 7.37]	11.3 [6.00, 26.7]	12.5 [3.43, 31.9]	11.8 [3.43, 31.9]
Missing	2 (50%)	0 (0%)	0 (0%)	2 (2%)
Co-administration, n (%)
Without P-gp inhibitors	4 (100%)	24 (77%)	68 (99%)	96 (92%)
With P-gp inhibitors	0 (0%)	7 (23%)	1 (1%)	8 (8%)

Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BSA, body surface area; CrCl, creatinine clearance; CREAT, creatinine; Max, maximum; Min, minimum; N, number of sample size; NVAF, nonvalvular atrial fibrillation; P-gp, P-glycoprotein; SD, standard deviation; TBIL, total bilirubin; WT, body weight.

External Validation of the ENGAGE Model

The ENGAGE PopPK model was externally validated using PK data from Chinese NVAF patients. Diagnostic plots revealed strong concordance between IPRED and observed concentrations. However, population predicted concentrations (PRED) exhibited poor alignment with observed values, with most PRED values underestimating the observations. The mean relative error percentage (PE%) for IPRED was −2% (within ± 10%), with a mean absolute relative error percentage (APE%) of 8% (below 25%). For PRED, the PE% was −29%, and the APE% reached 35% ([Supplementary Fig. S1], available in the online version only). These findings suggest limited predictive accuracy for PRED in the ENGAGE model. pc-VPC further indicated that observed concentrations fell outside the 2.5th, 50th, and 97.5th percentile ranges of simulated data ([Supplementary Fig. S2], available in the online version only), with dissimilar distribution patterns between observed concentrations and PRED. This underscores the ENGAGE model's inability to accurately reflect the PK trends and variability of edoxaban in Chinese NVAF patients.

PopPK Analysis

Given the limited applicability of the ENGAGE model (primarily developed using data from non-Asian populations) to Chinese NVAF patients, a new PopPK model was constructed using data from Chinese NVAF patients. The optimal model was a two-compartment structure with first-order absorption and linear elimination ([Fig. 1]). The base model incorporated allometric scaling with WT for CL/F, inter-compartmental clearance (Q/F), central compartment volume (Vc/F), and peripheral compartment volume (Vp/F). Exponents for Vc/F and Vp/F were fixed at 1, while the exponent for Kel was set to −0.25, and Q/F was fixed at 0.75. Inclusion of CrCl as a covariate for Kel significantly improved the OFV (ΔOFV = −12.54) and was retained. The relationship between Ka and Kel was parameterized using a flip-flop kinetics model. Inter-individual variability was modeled with an exponential error approach for Vc/F, Kel, and Q/F, while RUV was best described by combined additive and proportional error models.

Fig. 1 Schematic representation of the pharmacokinetic model. Following oral administration, edoxaban enters a depot compartment (representing the absorption site), from which it is absorbed into the central compartment via a first-order absorption process characterized by the absorption rate constant (Ka). The two-compartment model consists of the apparent distribution volume of the central compartment (Vc/F) and that of the peripheral compartment (Vp/F), which are interconnected by the intercompartmental clearance (Q). Drug elimination occurs from the central compartment through a first-order process described by the elimination rate constant (Kel). DOSE, administered oral dose; p.o., oral administration.

Covariate prescreening identified no significant correlations (p < 0.01) with potential covariates ([Supplementary Table S1], available in the online version only), resulting in none being retained in the final model. Final PopPK parameter estimates are presented in [Table 2].

Table 2 Population pharmacokinetic parameters of edoxaban

Pharmacokinetic parameter (unit)		Population mean	RSE%	95% Confidence interval
CL/F (L/h)	θ 1 · θ 2	21.168	17.0	19.1–23.28
Vc/F (L)	θ 1	147	3.9	136–159
Kel (1/h)	θ 2	0.144	3.0	0.136–0.153
Ka (1/h)	θ 1 + θ 2	5.624	87.6	4.14–7.09
Q/F (L/h)	θ 4	3.17	19.4	2.18–4.62
Vp/F (L)	θ 5	374	952.0	5.8–24,100
Effect of CrCl on Kel	θ 6	0.37	13.5	0.272–0.467
IIV_CL/F (%)	η 1·η 2	29.41	25.6	22.1–36.7
IIV_Vc/F (%)	η 1	25.9	10.0	20.1–30.8
IIV_Kel (%)	η 2	13.4	23.6	3.62–18.5
IIV_Q/F (%)	η 3	95	17.2	48.2–139
Proportional residual error (%)	ε 1	22	2.5	20.9–23.1
Additional residual error (ng/mL)	ε 2	2.48	21.7	0.955–3.38

Abbreviations: CL/F, apparent clearance; CrCl, creatinine clearance; IIV, inter-individual variability; Ka, absorption rate constant; Kel, elimination rate constant; Q/F, apparent inter-compartmental clearance; RSE, relative standard error; Vc/F, apparent distribution volume of central compartment; Vp/F, apparent distribution volume of peripheral compartment.

Model Evaluation

The GOF plots for the final PopPK model were evaluated graphically ([Fig. 2]). The scatter plots comparing observed concentrations with population and individual predictions demonstrated the model's robust fit across the observed range. The conditional weighted residuals exhibited no systematic trends against time or population predictions, affirming the model's lack of bias. Due to variable dosing, a pc-VPC was employed,[28] showing close alignment between the observed 2.5th, 50th, and 97.5th percentiles and simulated data across all postdose time intervals ([Fig. 3]).

Fig. 2 Goodness-of-fit plots. (A) Individual predicted concentration (IPRED) versus dependent variable (DV). DV, which indicated the observed concentration in NONMEM. (B) Population predicted concentrations (PRED) versus DV. (C) Time after dose versus conditional weighted residuals (CWRES). (D) PRED versus CWRES. The red lines represent the observed data trends, while the black lines show the expected trends. Blue circles denote the observed edoxaban concentrations in Chinese nonvalvular atrial fibrillation (NVAF) patients.

Fig. 3 Prediction-corrected visual predictive check for the final population pharmacokinetic model of edoxaban. The red line shows the median, and the blue lines show the 2.5th and 97.5th percentiles of observed edoxaban plasma concentrations. The shaded areas represent the 95% confidence intervals for the simulated edoxaban plasma concentrations at the 2.5th, 50th, and 97.5th percentiles. Blue circles represent the observed concentrations (corrected by prediction).

Model-Based Simulation

Key factors influencing edoxaban exposure were explored by simulating exposure across eight virtual patient groups, defined by typical PK parameters and variability under different dosing regimens ([Fig. 4], [Supplementary Fig. S3], available in the online version only). Patients in groups 3, 4, and 5 (30 mg q.d., dose-reduced for one or two factors) exhibited higher AUC_ss, C_max,ss, and C_avg,ss than group 2 (30 mg q.d., standard patients), but lower values than group 1 (60 mg q.d., standard patients), indicating that dose-adjusted patients experienced elevated exposure relative to nonadjusted patients receiving the same dose. C_min,ss was higher in group 5 (low WT and moderate renal impairment, 30 mg q.d.) compared with group 1, and the combination of low WT and moderate renal impairment further increased AUC_ss and C_avg,ss relative to groups 3 and 4 (single factors). Among 15 mg q.d. dose-adjusted patients (groups 6–8), C_min,ss was higher in group 8 (SRI and low WT) than in group 2, with AUC_ss and C_avg,ss trending upward compared to groups 6 and 7. These results suggest that exposure escalates with increasing dose-adjustment factors, even under the same dosing regimen. Additionally, the trough concentrations of edoxaban at different doses in elderly patients (80 years) with low WT (WT = 50 kg) and normal WT (WT = 70 kg) under varying renal function conditions were simulated ([Fig. 5]).

Fig. 4 Simulation of exposure measures. Boxplots are shown for the following subgroups: (1) WT = 70 kg, CrCl = 84.35 mL/minute, AGE = 65 year; (2) WT = 70 kg, CrCl = 84.35 mL/minute, AGE = 65 year; (3) WT = 70 kg, CrCl = 35 mL/minute, AGE = 65 year; (4) WT = 50 kg, CrCl = 84.35 mL/minute, AGE = 65 year; (5) WT = 50 kg, CrCl = 35 mL/minute, AGE = 65 year; (6) WT = 70 kg, CrCl = 20 mL/minute, AGE = 80 year; (7) WT = 70 kg, CrCl = 84.35 mL/minute, AGE = 80 year; (8) WT = 50 kg, CrCl = 20 mL/minute, AGE = 80 year. The horizontal black line inside the box represents the median. The bottom and top of the box represent the 25th and 75th percentiles, respectively. The whiskers extending from both ends of the box represent 1.5 times the interquartile range (IQR). Black circles represent outliers. AUC_ss, area under the plasma concentration-time curve; C_avg,ss, average plasma concentration; C_max,ss, maximum concentration; C_min,ss, trough concentration.

Fig. 5 Simulation of edoxaban plasma through concentration in elderly patients (80 years) with varying renal function conditions. The error bars signify 95% confidence intervals. The solid dots represent the median value of each treatment group. Each solid line represents the prediction from the logistic regression model. The dashed line indicates an edoxaban warning trough concentration of 22.7 ng/mL.[14] C_min,ss, trough concentration; CrCl, creatinine clearance; q.d., once daily.

Discussion

This study presents the first PopPK model tailored for edoxaban in Chinese NVAF patients. This model effectively characterized the PK profile of edoxaban in this population, identifying WT and CrCl as pivotal covariates influencing PK parameters. Our findings reveal that the ENGAGE PopPK model, predominantly derived from a non-Asian cohort, lacks direct applicability to Chinese NVAF patients. Notably, edoxaban exposure in this Chinese population exceeds that observed in the ENGAGE model's reference population ([Supplementary Table S2], available in the online version only).

The PopPK model developed herein for Chinese NVAF patients is a two-compartment model with first-order absorption, contrasting with the delayed absorption kinetics in the ENGAGE model. In the ENGAGE model, CL/F encompasses both renal and nonrenal clearance.[9] The divergence in absorption dynamics may stem from the ENGAGE model's reliance on dense sampling data from healthy subjects, versus the sparse sampling from Chinese NVAF patients in this study. For a typical patient with a CrCl of 84.35 mL/minute, our model estimates a CL/F of 21.17 L/hour, compared to 28.28 L/hour in the ENGAGE model and 24.09 L/hour in Japanese patients.[9] [17] This suggests a lower CL/F in Asian populations relative to non-Asian populations. Furthermore, the simulated AUC_ss in the ENGAGE model is 32% lower than those in this model ([Supplementary Table S2], available in the online version only), potentially attributable to the reduced CL/F in Chinese patients, which drives heightened edoxaban exposure.

Consistent with the ENGAGE model, our analysis confirms CrCl and WT as significant covariates positively correlated with CL/F. Prior clinical studies have reported that edoxaban exposure was elevated in East Asian patients with lower WT or compromised renal function.[21] [29] The population included in the ENGAGE model exhibited a median WT of 83 kg, with 81% White/Caucasian participants,[9] whereas our cohort had a median WT of 69.7 kg. This lower WT likely contributes to the reduced CL/F observed in our model relative to the ENGAGE model. We also assessed the influence of sex on edoxaban exposure, but its lack of statistical significance precluded inclusion in the final model.[30] Similarly, while concomitant use of P-gp inhibitors showed a trend toward increased exposure,[8] [31] the sample size of patients with coadministration with P-gp inhibitors was insufficient to demonstrate a statistically significant effect. Therefore, concomitant use of P-gp inhibitors was not included as a covariate in the final model.

Simulations revealed that patients receiving edoxaban 30 mg q.d. with moderate renal impairment (group 3), low WT (group 4), or both (group 5) exhibited significantly elevated exposure metrics (AUC_ss, C_max,ss, C_min,ss, and C_avg,ss) compared to standard patients on the same dose (group 2). Specifically, patients in group 5 demonstrated a 72% higher AUC_ss than group 2, 36% higher than group 3, and 35% higher than group 4. Despite doubling the dose to 60 mg q.d., AUC_ss of group 1 surpassed that of group 5 by only 16%, indicating that even with a halved dose, patients with these combined risk factors retain substantial exposure.

The inclusion of patients eligible for the 15 mg q.d. dose, characterized by advanced age, low WT, and renal impairment, likely influenced the overall PK parameters. Such traits are associated with altered drug metabolism and clearance, resulting in reduced CL/F and higher exposure. In the 15 mg q.d. cohort, group 8 (low WT, advanced age, and SRI) exhibited an AUC_ss 26 and 61% higher than groups 6 (SRI and advanced age) and 7 (low WT and advanced age), respectively, and 1% higher than group 2 (standard 30 mg q.d.). This suggests that multiple dose-reduction factors amplify exposure beyond that seen with single-factor adjustments, potentially heightening bleeding risk.

Salazar et al identified C_min,ss as the most reliable predictor of bleeding risk, reporting values of approximately 11.4 ng/mL for 30 mg q.d. and 22.7 ng/mL for 60 mg q.d., with levels exceeding 22.7 ng/mL markedly increasing bleeding likelihood, potentially surpassing the bleeding risk profile of warfarin.[14] [32] Our model predicts C_min,ss values of approximately 12 ng/mL for 30 mg q.d. and 20.6 ng/mL for 60 mg q.d. in Chinese NVAF patients. In group 5 (30 mg q.d., moderate renal impairment and low WT), C_min,ss was 28, 120, 21, and 100% higher than in groups 1, 2, 3, and 4, respectively, reaching 26.4 ng/mL. Despite dose halving from 60 to 30 mg, this elevation approximates or exceeds C_min,ss (60 mg q.d.) of group 1, highlighting the need for potential clinical monitoring in such patients due to heightened bleeding risk.

In 30 mg q.d. groups, group 3 with reduced CrCl alone had a C_min,ss of 21.9 ng/mL, which was higher than that (13.2 ng/mL) in group 4 with low WT alone. The same trends were also observed in the 15 mg q.d. elderly groups. These results suggest that when stratifying bleeding risk across various dosing regimens, renal function, represented by CrCl, should be prioritized over WT.

The ELDERCARE-AF study, assessing 15 mg edoxaban in elderly Japanese patients (≥80 years) with elevated bleeding risk, reported trough concentrations of 17.3 ± 13.9 ng/mL, aligning with our findings.[32] Despite different study designs, ELDERCARE-AF focused on high-risk Japanese patients, while our study encompassed a broader Chinese NVAF cohort; this similarity underscores consistent PK behavior in specific high-risk elderly subgroups. Furthermore, we simulated the trough concentrations of edoxaban at three different doses (15, 30, and 60 mg) in elderly (80 years old) NVAF patients with low body weight (WT = 50 kg) and normal body weight (WT = 70 kg) and different renal functions. With a warning concentration of 22.7 ng/mL,[14] we found that 60 mg of edoxaban seemed inappropriate as the therapeutic dose for patients aged 80 with low body weight. Elderly NVAF patients aged 80 years with normal weight, moderate and mild renal injury (30 mL/minute < CrCl < 80 mL/minute), and without other bleeding risk factors could be treated with 30 mg of edoxaban. The results support that edoxaban 15 mg as the therapeutic dose may be an appropriate regimen in very elderly (80-year-old) NVAF patients with low body weight and combined CrCl < 45 mL/minute.

The primary limitation of this study is its modest sample size (n = 104), potentially constraining the robustness of the PopPK model for Chinese NVAF patients. Moreover, the limited observation period and small sample size restricted the explanation of the association between clinical outcomes and edoxaban through concentration, so further research is needed.

Additionally, the limited number of patients receiving concomitant P-gp inhibitors precluded a comprehensive evaluation of their impact on edoxaban exposure. The study also included a few patients with extreme WT (<45 kg or >120 kg) or CrCl < 20 mL/minute, limiting generalizability to these subgroups. Furthermore, the final model developed may underpredict edoxaban concentrations at high concentrations, as demonstrated by the PRED-DV plot ([Fig. 2]). Future research should explore the PK profiles of edoxaban in patients with extreme WT and SRI to refine dosing strategies further.

Conclusion

This study established the first PopPK model for edoxaban in Chinese NVAF patients. The model identified WT and CrCl as key covariates influencing CL/F of edoxaban, crucial for modulating drug exposure. Simulations revealed that patients with multiple clinical risk factors necessitating a 50% dose reduction exhibited heightened edoxaban exposure compared to those with a single factor. Similarly, within the 15 mg dosing cohort, an increase in clinical risk factors leading to dose reduction also resulted in higher overall edoxaban exposure. Notably, the C_min,ss for patients with both moderate renal impairment and low WT who received the 30 mg q.d. dose was higher than that of patients receiving the 60 mg q.d. dose. Given the robust linear association between edoxaban trough concentration and bleeding risk, these patients may require clinical monitoring for hemorrhage potential.

Contributors' Statement

Y.D. and S.Z. collected detailed patient data. S.D. performed experiments and wrote the manuscript. J.J. and L.C. performed experiments. Z.L., X.Q., L.C., and S.H. performed data acquisition. Y.G., H.C., and S.S. analyzed the data. L.H. and L.T. conceived and supervised experiments, edited the manuscript, and secured funding for this study. All authors reviewed, edited, and approved the manuscript.

Conflict of Interest

The authors declare that they have no conflict of interest.

Acknowledgment

The authors thank the participants, their families, and all study site staff for their contributions. The graphical abstract was created with BioRender.com (https://BioRender.com/rq965wo).

Data Availability Statement

The data that support the findings of the study are available from the corresponding author upon reasonable request (L.H: [e-mail] and L.T.: [e-mail]). The full-text version of this article contains a data supplement.

^‡ These authors contributed equally to this article.

• References

• 1 Hindricks G, Potpara T, Dagres N. et al; ESC Scientific Document Group. 2020 ESC guidelines for the diagnosis and management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS): the task force for the diagnosis and management of atrial fibrillation of the European Society of Cardiology (ESC) developed with the special contribution of the European Heart Rhythm Association (EHRA) of the ESC. Eur Heart J 2021; 42 (05) 373-498
• 2 Bennaghmouch N, de Veer AJWM, Bode K. et al. Efficacy and safety of the use of non-vitamin K antagonist oral anticoagulants in patients with nonvalvular atrial fibrillation and concomitant aspirin therapy: a meta-analysis of randomized trials. Circulation 2018; 137 (11) 1117-1129
• 3 Steffel J, Collins R, Antz M. et al; External reviewers. 2021 European Heart Rhythm Association practical guide on the use of non-vitamin K antagonist oral anticoagulants in patients with atrial fibrillation. Europace 2021; 23 (10) 1612-1676
• 4 Ogata K, Mendell-Harary J, Tachibana M. et al. Clinical safety, tolerability, pharmacokinetics, and pharmacodynamics of the novel factor Xa inhibitor edoxaban in healthy volunteers. J Clin Pharmacol 2010; 50 (07) 743-753
• 5 Chen X, Liu D, Wu Y. et al. A single and multiple postprandial dose study investigating the pharmacokinetics and pharmacodynamics of edoxaban in healthy Chinese volunteers
. Int J Clin Pharmacol Ther 2017; 55 (03) 256-263
• 6 Mikkaichi T, Yoshigae Y, Masumoto H. et al. Edoxaban transport via P-glycoprotein is a key factor for the drug's disposition. Drug Metab Dispos 2014; 42 (04) 520-528
• 7 Kato T, Mikkaichi T, Yoshigae Y. et al. Quantitative analysis of an impact of P-glycoprotein on edoxaban's disposition using a human physiologically based pharmacokinetic (PBPK) model. Int J Pharm 2021; 597: 120349
• 8 Mendell J, Zahir H, Matsushima N. et al. Drug-drug interaction studies of cardiovascular drugs involving P-glycoprotein, an efflux transporter, on the pharmacokinetics of edoxaban, an oral factor Xa inhibitor. Am J Cardiovasc Drugs 2013; 13 (05) 331-342
• 9 Krekels EH, Niebecker R, Karlsson MO. et al. Population pharmacokinetics of edoxaban in patients with non-valvular atrial fibrillation in the ENGAGE AF-TIMI 48 study, a phase III clinical trial. Clin Pharmacokinet 2016; 55 (09) 1079-1090
• 10 Yamashita T, Koretsune Y, Yasaka M. et al. Randomized, multicenter, warfarin-controlled phase II study of edoxaban in Japanese patients with non-valvular atrial fibrillation. Circ J 2012; 76 (08) 1840-1847
• 11 Ruff CT, Giugliano RP, Antman EM. et al. Evaluation of the novel factor Xa inhibitor edoxaban compared with warfarin in patients with atrial fibrillation: design and rationale for the effective anticoagulation with factor Xa next generation in atrial fibrillation-thrombolysis in myocardial infarction study 48 (ENGAGE AF-TIMI 48). Am Heart J 2010; 160 (04) 635-641
• 12 Giugliano RP, Ruff CT, Braunwald E. et al; ENGAGE AF-TIMI 48 Investigators. Edoxaban versus warfarin in patients with atrial fibrillation. N Engl J Med 2013; 369 (22) 2093-2104
• 13 Matsushima N, Lee F, Sato T, Weiss D, Mendell J. Bioavailability and safety of the factor Xa inhibitor edoxaban and the effects of quinidine in healthy subjects. Clin Pharmacol Drug Dev 2013; 2 (04) 358-366
• 14 Salazar DE, Mendell J, Kastrissios H. et al. Modelling and simulation of edoxaban exposure and response relationships in patients with atrial fibrillation. Thromb Haemost 2012; 107 (05) 925-936
• 15 Yin OQ, Tetsuya K, Miller R. Edoxaban population pharmacokinetics and exposure-response analysis in patients with non-valvular atrial fibrillation. Eur J Clin Pharmacol 2014; 70 (11) 1339-1351
• 16 Yamashita T, Igawa Y, Fukuzawa M, Hayashi T, Hennig S, Okumura K. Pharmacokinetics of edoxaban 15 mg in very elderly patients with nonvalvular atrial fibrillation: a subanalysis of the ELDERCARE-AF study. Thromb Haemost 2024; 124 (09) 874-882
• 17 Shimizu T, Tachibana M, Kimura T, Kumakura T, Yoshihara K. Population pharmacokinetics of edoxaban in Japanese atrial fibrillation patients with severe renal impairment. Clin Pharmacol Drug Dev 2017; 6 (05) 484-491
• 18 Niebecker R, Jönsson S, Karlsson MO. et al. Population pharmacokinetics of edoxaban in patients with symptomatic deep-vein thrombosis and/or pulmonary embolism–the Hokusai-VTE phase 3 study. Br J Clin Pharmacol 2015; 80 (06) 1374-1387
• 19 January CT, Wann LS, Calkins H. et al. 2019 AHA/ACC/HRS focused update of the 2014 AHA/ACC/HRS guideline for the management of patients with atrial fibrillation: a report of the American College of Cardiology/American Heart Association Task Force on clinical practice guidelines and the Heart Rhythm Society in collaboration with the Society of Thoracic Surgeons. Circulation 2019; 140 (02) e125-e151
• 20 Clapham RE, Speed V, Byrne R. et al. Can edoxaban be used at extremes of bodyweight and in patients with a creatinine clearance ≥95 mL/min? - a population pharmacokinetic analysis. Thromb Res 2024; 242: 109118
• 21 Boriani G, De Caterina R, Manu MC, Souza J, Pecen L, Kirchhof P. Impact of weight on clinical outcomes of edoxaban therapy in atrial fibrillation patients included in the ETNA-AF-Europe registry. J Clin Med 2021; 10 (13) 2879
• 22 Lip GY, Agnelli G. Edoxaban: a focused review of its clinical pharmacology. Eur Heart J 2014; 35 (28) 1844-1855
• 23 Elenjickal EJ, Travlos CK, Marques P, Mavrakanas TA. Anticoagulation in patients with chronic kidney disease. Am J Nephrol 2024; 55 (02) 146-164
• 24 Koretsune Y, Yamashita T, Kimura T, Fukuzawa M, Abe K, Yasaka M. Short-term safety and plasma concentrations of edoxaban in Japanese patients with non-valvular atrial fibrillation and severe renal impairment. Circ J 2015; 79 (07) 1486-1495
• 25 Jönsson S, Simonsson US, Miller R, Karlsson MO. Population pharmacokinetics of edoxaban and its main metabolite in a dedicated renal impairment study. J Clin Pharmacol 2015; 55 (11) 1268-1279
• 26 Parasrampuria DA, Marbury T, Matsushima N. et al. Pharmacokinetics, safety, and tolerability of edoxaban in end-stage renal disease subjects undergoing haemodialysis. Thromb Haemost 2015; 113 (04) 719-727
• 27 Okumura K, Akao M, Yoshida T. et al; ELDERCARE-AF Committees and Investigators. Low-dose edoxaban in very elderly patients with atrial fibrillation. N Engl J Med 2020; 383 (18) 1735-1745
• 28 Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction-corrected visual predictive checks for diagnosing nonlinear mixed-effects models. AAPS J 2011; 13 (02) 143-151
• 29 Yamashita T, Koretsune Y, Yang Y. et al. Edoxaban vs. Warfarin in East Asian patients with atrial fibrillation-an ENGAGE AF-TIMI 48 subanalysis. Circ J 2016; 80 (04) 860-869
• 30 Yin OQ, Miller R. Population pharmacokinetics and dose-exposure proportionality of edoxaban in healthy volunteers. Clin Drug Investig 2014; 34 (10) 743-752
• 31 Yoshioka H, Sato H, Hatakeyama H, Hisaka A. Model-based meta-analysis to evaluate optimal doses of direct oral factor Xa inhibitors in atrial fibrillation patients. Blood Adv 2018; 2 (10) 1066-1075
• 32 de Vries TAC, Mallick IU, Bhagirath VC. et al. Usual on-therapy ranges of drug concentrations in patients with atrial fibrillation treated with direct oral anticoagulants: a systematic review and meta-analysis. Thromb Haemost 2024; 125 (06) 563-573

Publication History

Received: 11 June 2025

Accepted: 31 January 2026

Article published online:
12 February 2026