Pharmacokinetic/Pharmacodynamic Modelling and Simulation of Lusutrombopag, a Novel Thrombopoietin Receptor Agonist, for the Treatment of Thrombocytopenia in Patients with Chronic Liver Disease Undergoing Invasive Procedures

For PK modelling, 4196 plasma lusutrombopag concentration data from 427 subjects (78 healthy subjects and 349 thrombocytopenic CLD patients) were used. For PK/PD modelling, 3526 platelet counts from 347 of the above 349 patients were used. The study design is summarized in electronic supplementary Table S1. The 46 samples with plasma concentrations below the lower limit of quantification (LLOQ) after the initial dosing were treated as missing in the analyses. Four plasma concentrations from one patient were excluded due to unidentified sampling times, and 17 plasma concentrations taken on day 10 from 17 subjects in a multiple-dose study were also excluded because these concentrations were anomalously higher, as previously reported [9]. The 20 platelet counts from two patients were excluded because these platelet count profiles were anomalous due to concomitant treatment with interactive drugs.

All clinical studies were conducted according to the principles of the Declaration of Helsinki and in accordance with the Guideline of Good Clinical Practice. These studies were approved by independent Ethics Committees, and signed informed consent forms were obtained from all subjects.

Plasma concentrations of lusutrombopag were measured by a validated high-performance liquid chromatography-tandem mass spectrometry with an LLOQ of 0.100 ng/mL (1.00 ng/mL only in Study 0713M0611). The precision of the bioanalytical method was 0.8–12.7%.

A three-compartment model with the first-order absorption and lag time, which was the same structural model as in healthy subjects [9], was used as a PK starting base model. PK parameters were estimated from data in both healthy subjects and patients with CLD. Since the bioavailability of lusutrombopag was dependent on formulations and food conditions [9], a relative bioavailability, F1, was parameterized for the formulation (solution, 0.25 mg, 1 mg, 2 mg, or 3 mg tablet) in the food condition (fasted or fed). F1 in the fasted state of the solution was set to 1 as a reference value. F1 values in the fed state of the solution, the 1 mg tablet, and the 3 mg tablet were estimated. Based on the results in the relative bioavailability/food effect studies, F1 in the fed state of the 2 mg tablet was fixed at 0.857, and F1 in the fed state of the 0.25 mg tablet was calculated from 0.820 multiplied by F1 in the fed state of the solution. The resulting F1 for the dose regimen used was determined based on F1 of the formulation in the food condition, considering the proportions of different tablets used. In addition, as a visual inspection suggests, different plasma concentration profiles between the solution in the fed state and the other conditions, the parameters related to absorption (i.e. first-order absorption rate constant [KA] and lag time for absorption [ALAG1]) for the solution in the fed state were estimated separately. The interindividual variability (IIV) was assumed to be log-normally distributed for apparent total clearance (CL/F), apparent volume of distribution in the central compartment (V2/F) or peripheral compartment (V3/F and V4/F), apparent intercompartmental clearance (Q3/F and Q4/F), and KA. A proportional residual error model and a combination residual error model (proportional residual error + additive residual error) were tested. In the covariate modelling, the following effects were tested: age, body weight, sex, Child–Pugh classification (A, B or C) [13], creatinine clearance estimated using the Cockcroft–Gault equation [14], ethnicity (Japanese or non-Japanese subjects), and subject population (healthy subjects or CLD patients) on CL/F; age, body weight, sex, ethnicity, and subject population on V2/F; and body weight on V3/F. Covariates were normalized to the median values for the continuous covariates using the following equation (Eq. 1):

where θ_i is the individual predicted parameter for the individual continuous covariate value COV_i, θ_pop is a population mean parameter, θ_COV is an estimate for the covariate effect, and COV_med is the median value of the covariate.

The following equation was used for the categorical covariates (Eq. 2):

where CAT_i is the individual categorical variable for sex, ethnicity, and subject population with either 0 or 1 (0 for male/non-Japanese/healthy subjects, and 1 for female/Japanese/CLD patients), θ_{CAT_i = 0} is the typical values of the parameter for the individual categorical variable of 0, and θ_{CAT_i = 1} is the covariate effect for the individual categorical variable of 1.

Covariate models were constructed by a combination of screening for inclusion and stepwise backward deletion based on the Chi-square test. The significance level of 0.05 for covariate models versus the base model was used for screening (the difference in the value of the objective function [ΔOBJ] of less than − 3.84 for one degree of freedom was significant). The significant covariates at screening were included in the base model to construct a full model. The backward deletion with the significance level of 0.01 between models with and without the covariate was used to construct a final model (the deletion of the covariate with an ΔOBJ of < 6.64 for one degree of freedom was accepted).

A semi-physiological model incorporating the platelet production process in bone marrow, and the degradation process in circulating blood, was used as a PD structural model, composed of transit and platelet compartments [9]. In the model, lusutrombopag stimulates the zero-order production rate of platelet precursors in a manner dependent of plasma lusutrombopag concentrations. Four- to six-compartment models (three- to five-transit and one-platelet compartment models) were tested and compared based on the objective function values. A five-compartment model (four-transit and one-platelet compartment model) was selected, as shown in Eqs. 3–7.

where KPR is a zero-order rate for the production of platelet precursors, KM is a first-order rate constant for the maturation of platelet precursors, and KL is a first-order rate constant for the elimination of platelets. P1–P4 are platelet precursor counts, and PLT is the observation variable of platelet counts. KM, KL, and PLT0 were estimated from data in the CLD patients. The KPR was calculated as PLT0 × KL, assuming a steady state of platelet count. The initial conditions at steady state were: P1(0)–P4(0) = KPR/KM, PLT(0) = PLT0.

The linear model shown in Eq. 8 was tested to explain the drug effect (E in Eq. 3).

where SLOP is the slope relating to plasma lusutrombopag concentrations, and C is the plasma lusutrombopag concentration. As a saturable manner did not appear in the observation of exposure–response relationship, a sigmoid maximum effect (E_max) model was not tested. A first-order rate constant for the production [15] and a feedback function [9, 15] were tested.

IIV for KM, KL, SLOP, and PLT0 was assumed to be log-normally distributed. A proportional residual error model and a combination residual error model were tested.

In the covariate modelling, effects of age, body weight, Child–Pugh class (A, B or C) or numerical score, sex, ethnicity, and observed baseline platelet count on SLOP were tested by screening, followed by the stepwise backward deletion similar to the PK modelling.

The developed PK/PD model was employed to assess dose response, necessity of platelet monitoring, and effect of body weight on platelet counts. For the dose response, 2, 3 and 4 mg once-daily for 7 days were tested. For the necessity of monitoring platelet counts during treatment, the following treatment completion criterion was employed: if platelet counts were ≥ 50,000/µL and an increase from the baseline was ≥ 20,000/µL, administration of lusutrombopag was stopped. The Monte-Carlo simulations were performed with or without the treatment completion criterion based on platelet counts prior to the dose on day 6 (1 day) or days 5–7 (3 days). For each simulation scenario, 200-replication simulations were performed by generating 69,400 virtual subjects with resampling of the covariate data from the analysis dataset. For the effect of body weight, 2000 simulations (2000 virtual patients per group), using a 3 mg once-daily dose for 7 days, were performed by body weight group according to the uniform distribution of body weight, Child–Pugh score <9, and male to female ratio of 1:1 for non-Japanese CLD patients. PLT0 was simulated according to a log-normal distribution based on the estimate (mean 39,200/µL and coefficient of variation [CV] 26.5%). The simulated platelet counts using PLT0 of > 50,000/µL were excluded from the summary. Platelet counts of virtual subjects were simulated every 24 h from 0 to 696 h. The following platelet metrics were calculated from the simulated platelet counts for each simulation scenario: (1) 5th, 50th and 95th percentiles of peak platelet count; (2) as an efficacy index, probability to attain 50,000/µL of platelet count from days 9 to 14, which is the period when invasive procedures are expected to be scheduled; and (3) as a safety index, probability to exceed 200,000/µL of platelet count on any time points.

NONMEM version 7.3 (ICON Development Solutions, Ellicott City, MD, USA) was used [18]. Perl-speaks NONMEM version 4.2.0 was used to execute a NONMEM run for parameter estimations and simulations, VPC, and a non-parametric bootstrap method for calculating 95% CIs of parameter estimates [19], and R version 3.0.2 was used for summarization and the graphics [20].

Demographic data are presented in Table 1. Figure 1 shows the scheme of the final PK/PD model, and Table 2 shows the population PK/PD parameters of the final model. The relative standard error for parameter estimates was ≤ 23.0%. The median values of estimates based on the bootstrap approach were close to point estimates for all parameters and the CIs did not include a null value.

For the PK data, the three-compartment model with the first-order absorption and lag time, and the proportional residual error model, were selected. The IIV for Q3/F, Q4/F and V4/F was excluded because it was close to zero. The dose-normalized VPC indicated the predictions of the final PK model described the observed PK profiles in healthy subjects and thrombocytopenic CLD patients (Fig. 2). The following covariates were retained in the final model: body weight on CL/F, V2/F and V3/F; sex and ethnicity (Japanese/non-Japanese) on CL/F; and subject population (healthy subjects/thrombocytopenic CLD patients) on CL/F and V2/F. The CI from the bootstrap for the effect of sex and ethnicity on CL/F was within 0.80–1.25 (electronic supplementary Fig. S2). The CL/F ratio was 0.870 (95% CI 0.788–0.952) in CLD patients relative to healthy subjects. As the lower limit of the 95% CI was close to 0.80, the difference in CL/F between the subject populations was not considered clinically significant. The V2/F in CLD patients was 1.46-fold (95% CI 1.25–2.17) higher than that in healthy subjects.

For the PD data, the five-compartment model (four-transit compartment and one-platelet compartment models) with the zero-order rate for production was selected as a base model. The drug effect was well-described with the linear model. The KL in the PD model with feedback function was estimated to be 0.00444/h, which was lower than in healthy subjects (0.00648/h) [9]. As the feedback is a regulation function on production in case of excessive platelets, the feedback function was not applied. The prediction-corrected VPC indicated the predictions of the final PK/PD model described the platelet counts in thrombocytopenic CLD patients (Fig. 3). In the covariate modelling, SLOP increased for Child–Pugh score of ≥ 9, while no clear relationships were observed between SLOP and age, body weight, ethnicity, sex, or observed baseline platelet counts (electronic supplementary Fig. S3).

The steady-state daily AUC and peak platelet counts were predicted based on empirical Bayes-estimated parameters following a 3 mg once-daily dose for 7 days in thrombocytopenic CLD patients (Fig. 4). The AUC of lusutrombopag decreased with increasing body weight. Furthermore, the AUC was 1.3-fold higher in Japanese patients than in non-Japanese patients. The median peak platelet counts were higher in the lowest body weight group (35 to < 45 kg) than in the other groups (45–130 kg), while the individual peak platelet counts overlapped among the body weight groups. Peak platelet counts in Japanese patients were similar to those in non-Japanese patients. Median AUC and peak platelet counts in patients with Child–Pugh class C were lower than those in patients with Child–Pugh class A or B, but the AUC was comparable with that in healthy subjects. The individual AUC and peak platelet counts overlapped among Child–Pugh classes.

Figure 5 shows simulated platelet metrics for 2, 3 and 4 mg once-daily doses for 7 days. The peak platelet counts were slightly different among the dose regimens. The probabilities to attain 50,000/μL on days 9–14 were 75.9%, 78.0% and 70.7% at the 2 mg dose; 85.2%, 87.1% and 80.8% at the 3 mg dose; and 90.4%, 92.0%, and 86.6% at the 4 mg dose, in all patients, Japanese and non-Japanese patients, respectively. The probabilities to exceed 200,000/μL were 0.07%, 0.09% and 0.02% at the 2 mg dose; 0.39%, 0.50% and 0.13% at the 3 mg dose; and 1.05%, 1.34% and 0.33% at the 4 mg dose, in all patients, Japanese patients and non-Japanese patients, respectively.

Figure 6 shows the simulated platelet metrics at 3 mg once daily for 7 days with or without the treatment completion criterion. Median peak platelet counts were similar between groups with or without the treatment completion criterion in any patient population. The probabilities of exceeding 200,000/μL in all patients, Japanese patients and non-Japanese patients, respectively, were 1.20%, 1.52% and 0.43% without the treatment completion criterion; 0.59%, 0.74% and 0.20% with the treatment completion criterion on day 6; and 0.39%, 0.50% and 0.13% with the treatment completion criterion on days 5–7. The probabilities of attaining 50,000/μL were similar between groups with and without the treatment completion criterion in any patient population. The probability for ≥ 50,000/µL was independent of days for treatment completion criterion since dose was stopped for patients who achieved 50,000/µL.

Figure 7 shows the simulated platelet metrics at a 3 mg once-daily dose for 7 days, by body weight group. Differences in median peak platelet counts were 3900–12,000/μL in each body weight group relative to the reference group (60 to < 80 kg). The probabilities of attaining 50,000/μL and exceeding 200,000/μL were 90.8% and 2.05% in the lowest body weight groups (35 to < 45 kg), respectively, and 69.5% and 0.12% in the highest body weight group (100–130 kg), respectively. The simulated platelet metrics between 3 and 4 mg in heavier patients (≥ 80 kg of body weight) indicated the differences were ≤ 6300/µL in median peak platelet counts and ≤ 8% in probability for ≥ 50,000/µL (electronic supplementary Fig. S4). It is suggested the 4 mg dose would provide the limited platelet increase in the heavier patients.