Population Pharmacokinetic and Pharmacodynamic Analysis of Linezolid and a Hematologic Side Effect, Thrombocytopenia, in Japanese Patients^,

Abstract

Linezolid is an antimicrobial agent to treat infections by Gram-positive pathogens, including methicillin-resistant Staphylococcus aureus (MRSA). While effective, linezolid treatment frequently is associated with hematological side effects, especially thrombocytopenia. However, little is known about the mechanism of this side effect and the exposure-response relationship. The present population pharmacokinetic/pharmacodynamic (PPK/PD) study was undertaken to elucidate the factors that determine linezolid levels, the relationship between exposure to linezolid and a decrease in platelet counts, and appropriate dosage adjustments based on exposure levels. In total, 50 patients (135 plasma samples) were used for the PPK analysis. The PPK analysis revealed that renal function and severe liver cirrhosis (Child Pugh grade C) significantly affect the pharmacokinetics of linezolid according to the equation clearance (liter/h) = 2.85 × (creatinine clearance/60.9)^0.618 × 0.472^CIR (CIR indicates cirrhosis status; 0 for noncirrhosis, 1 for cirrhosis patients). Using 603 platelet counts from 45 patients, a PPK/PD analysis with a semimechanistic pharmacodynamic model described the relationship between linezolid exposure and platelet counts quantitatively, and the newly constructed model was validated using external data (776 platelet counts from 60 patients). Simulation indicated considerable risks in patients with insufficient renal function (creatinine clearance, ≤30 ml/min) or severe liver cirrhosis. For these patients, a reduced dosage (600 mg/day) would be recommended for sufficient efficacy (area under the concentration-time curve over 24 h in the steady state divided by the MIC, >100) and safety.

INTRODUCTION

Linezolid (LZD), an antimicrobial agent, belongs to a family of oxazolidinones that inhibit bacterial protein synthesis by preventing the formation of 30S-50S ribosomal subunits and the translation initiation complex (10). LZD exhibits a broad spectrum of activity against Gram-positive pathogens, including vancomycin-resistant Enterococcus faecium (VRE) and methicillin-resistant Staphylococcus aureus (MRSA) (23). LZD has 100% oral bioavailability, reaches high concentrations at different sites, and is at least as potent as vancomycin for treating MRSA infections (19). Hence, LZD is administered to patients with serious MRSA infections who respond poorly to or cannot tolerate glycopeptides (6).

The major concern with the clinical use of LZD is hematological toxicity, especially thrombocytopenia, which can lead to a discontinuation of treatment and the need for a platelet transfusion (3, 16). The incidence of LZD-induced thrombocytopenia varies from 2.4 to 64.7% (3, 6, 14, 22, 28, 30, 34), and little is known about the mechanism responsible. Recent studies, however, have shown the incidence of LZD-induced thrombocytopenia to be significantly higher in patients treated long term (>14 days) (16) or with insufficient renal function (22, 25, 34), and it is lower in patients coadministered rifampin, which significantly reduced the area under the curve (AUC) of LZD (34, 12). Given these findings, the development of thrombocytopenia may depend on the degree of exposure to LZD; however, few studies have examined the relationship between exposure to LZD and the incidence of thrombocytopenia quantitatively. Forrest et al. showed that the LZD AUC and its treatment duration were highly associated with the decrease of platelet counts by the population pharmacokinetic/pharmacodynamic (PK/PD) modeling approach using data obtained during its clinical development (13).

Moreover, the need to adjust the dosage of LZD based on exposure has not yet been fully discussed. Although several pharmacokinetic studies, including population pharmacokinetic studies, have been conducted already (1, 26, 27, 29, 39), covariates to determine levels of LZD exposure have not been identified. Abe et al. performed a population pharmacokinetic study of Japanese infectious patients and revealed that the age and body weight significantly affected LZD clearance (CL) (1). However, in their report, the goodness-of-fit plots of population predictions showed serious underestimation, particularly at high plasma concentration levels, suggesting that they missed influential covariates that decrease LZD CL. Actually, Tsuji et al. reported that concentrations predicted using population parameters provided by Abe et al. deviated considerably less than actual observations with statistical significance (P < 0.01) in Japanese patients (38).

Population pharmacokinetic/pharmacodynamic (PPK/PD) analysis with nonlinear mixed-effect modeling is a conventional method that permits the development of an appropriate exposure-response model and the assessment of the influence of covariates on the overall PPK/PD profile of a drug (32). In the present study, a PPK/PD analysis was performed to (i) identify the factors influencing the pharmacokinetics of LZD and LZD-induced thrombocytopenia, (ii) elucidate the relationship between exposure to LZD and decreases in platelet count, and (iii) verify the need for dosage adjustments based on LZD levels.

MATERIALS AND METHODS

Study design and patients.

This was a prospective, open-labeled, uncontrolled study investigating the pharmacokinetics of LZD in Japanese patients with infectious disease and its relationship with the incidence of thrombocytopenia in Tottori University Hospital (Yonago, Japan). The study protocol was approved by the Tottori University Ethics Committee, and written informed consent was obtained from all individuals. From June 2006 to July 2010, patients who received linezolid (Zyvox; Pfizer, Kalamazoo, MI) were enrolled. The background of the patients, including laboratory tests and prescribed drugs, was collected through an electronic medical database available in the hospital. Creatinine clearance (CL_CR) was calculated by the Cockcroft-Gault equation (8). Patients with poor drug compliance or taking drugs known to interact with LZD (including rifampin) were excluded. In the assessment of platelet counts, patients who showed severe thrombocytopenia (<50 × 10³/μl) before LZD treatment, bleeding, receiving platelet transfusion, or taking any anticancer drugs during the LZD treatment also were excluded.

Drug administration and pharmacokinetic sampling.

All patients received 300 to 600 mg of LZD twice daily orally and/or by constant intravenous (i.v.) infusion within 1 to 2 h. Pharmacokinetic sampling was performed when a steady-state LZD concentration had been achieved (i.e., at least 3 days from the start of treatment). Blood samples were obtained at one to five points per patient just before the administration and 1.5, 2, 4, 5, 8, and 9 h after the administration. Plasma was separated by centrifugation for 15 min at 3,000 rpm and stored at −80°C until the analysis. Plasma concentrations of LZD were determined using high-performance liquid chromatography according to previous methods (36). Authentic LZD was kindly provided by Pfizer. The lower limit of quantification was 0.1 μg/ml.

Platelet change.

Platelet counts for each patient were made before, during, and after (1 to 20 days) the LZD treatment and were obtained through the electronic medical database. The samplings for measuring platelet counts were performed just before breakfast in the morning. Thrombocytopenia was defined as a decrease in the platelet count to <100 × 10³/μl.

Data analysis.

The population pharmacokinetic/pharmacodynamic (PPK/PD) analysis was performed using NONMEM, version VI, level 1.0 (Icon Development Solutions, Ellicott City, MD) with the first-order conditional estimation with interaction method (FOCE-INTER). The graphic processing of the NONMEM output was performed with Xpose (version 4.0, release 3) for the statistical package R (version 2.6.2) and S-PLUS (version 6.1; Mathematical Systems, Inc., Tokyo, Japan) (18). The PPK/PD model was built sequentially: first the PPK model was developed and then the time course of the platelet count was modeled, with each individual empirical Bayesian estimated (EBE) PK parameter calculated with the NONMEM POSTHOC option.

Population pharmacokinetic analysis.

A one-compartment model was used for the pharmacokinetic modeling. The absorption rate constant (k_a) was fixed at the value reported by Abe et al. (1) because of a lack of sampling points. The exponential error model and additive error model were used to describe the interindividual and residual variability, respectively. Body weight (BW), age, serum creatinine, CL_CR, total bilirubin, and liver cirrhosis (CIR) (Child Pugh grade C) were selected as candidates for the pharmacokinetic covariate. Covariate selection was performed as follows. First, the effect of each covariate on a parameter was screened using the generalized additive model (GAM) approach implemented in Xpose. Covariates selected during the GAM analysis then were further assessed based on the difference in the objective function value (OFV) estimated by NONMEM between hierarchical models. Forward inclusion and backward elimination were used to develop the covariate model. The significance levels for forward inclusion and backward elimination were set to 0.05 and 0.01, respectively. The adequacy of the constructed PPK model was assessed by goodness-of-fit (GOF) plots at each step during the model's development. GOF was investigated using the plots of observations versus population predictions (PRED) and individual prediction (IPRED), conditional weighted residuals (CWRES) versus time after administration (17), CWRES versus PRED, and absolute individual weighted residuals (|IWRES|) versus IPRED. To assess the robustness of the estimated parameter, a bootstrap analysis and case deletion diagnostics were performed with Perl-speaks-NONMEM (21).

Population pharmacokinetic/pharmacodynamic modeling.

The semimechanistic PK/PD model reported by Friberg et al. (15) was used to describe platelet count profiles with individual EBE PK parameters. In short, it consisted of a system-dependent part and a drug-dependent part. The system-dependent part resembled, in a simplified manner, the underlying physiological processes determining the platelet count in the circulation: (i) proliferation at the progenitor cell compartment; (ii) maturation, represented in the model by three transit compartments; (iii) degradation; and (iv) homeostatic regulation. Steps i to iv can be described by the first-order rate constant of proliferation of progenitor cells (k_prol), the mean transit time (MTT), the first-order rate constant of degradation (k_circ), and the feedback parameter (γ). Circ₀ represents the platelet count at baseline. We used the model k_prol = k_circ = k_tr, where k_tr is the first-order rate constant governing the transfer of immature cells between the transit compartments computed as (n + 1)/MTT, with n being the number of transit compartments included in the model. Before drug treatment, the initial conditions in the progenitor cell, transit, and circulation compartments are the same and are equal to Circ₀. The drug effect was represented by the parameter Slope and is included in the model as follows: k_prol × (1 − Slope × C_p), where C_p was the predicted LZD plasma concentration. This semimechanistic PK/PD model is schematically represented in Fig. 1. The exponential error model (for Circ₀) and the Box-Cox transformed distribution error model (for Slope and MTT) were used to describe the interindividual variability with the equation η_i = [(e^η_n)^Shape − 1]/Shape, where η_i is the Box-Cox-transformed random effect, η_n is a normally distributed random effect, and Shape is the parameter that determines the shape of the distribution. An exponential and additive combined error model was used to describe residual variability. The exploration of covariates was performed as in the PPK analysis. The adequacy of the PPK/PD model was assessed using GOF plots.

Fig. 1.

Model used to describe linezolid pharmacokinetics and platelet profile. k_a, absorption rate constant; V, volume of distribution; CL, clearance; C_p, plasma concentration of linezolid; k_prol, proliferation rate constant of progenitor cells; k_tr, transition rate constant; k_circ, degradation rate constant of circulating platelet; MTT, mean transit time; Circ₀, platelet count at baseline; γ, feedback parameter.

The model was further evaluated with PD validation data as follows: individual platelet profiles during LZD treatment were simulated by the PPK/PD model with actual dosage schedules and patients' backgrounds, including baseline platelet counts. After this step, (i) their distribution at each LZD treatment duration and LZD exposure level, (ii) the value at the nadir, and (iii) the percentage of patients showing thrombocytopenia (<100 × 10³/μl) were calculated and compared to the same descriptors from the actual data.

Model-based simulation for dosage adjustment.

Using the PPK/PD model, the plasma concentration and platelet profile of patients with a CL_CR of 100, 50, 30, or 10 ml/min and Child Pugh grade C cirrhosis (1,000 individuals in each) were simulated. The dosing schedules were set to 600 or 300 mg twice daily by 1 h of intravenous infusion. The simulated patient's background data were obtained by the random sampling of the current whole actual data. Random sampling and Monte Carlo simulations were performed by S-PLUS and NONMEM, respectively. For the assessment of efficacy, areas under the concentration-time curve over 24 h in the steady state divided by the MIC (AUC/MIC ratios) were computed in the steady-state condition in individuals, and the percentage of patients that achieved an AUC/MIC ratio of >100 was calculated for each group. The MIC was set to 2.0 μg/ml according to a previous report (2). For the assessment of safety, platelet counts at 7, 14, and 21 days after the start of LZD treatment were computed for individuals, and the rates of thrombocytopenia (<100 × 10³/μl) were calculated for each group.

RESULTS

A total of 110 patients were enrolled. Among them, PK samples were obtained from 50 patients. Platelet count profiles were obtained from all patients. The demographics and characteristics of each data set are summarized in Table 1.

Table 1.

Summary of patient characteristics^d

Parameter	PPK/PD model development data	PPK/PD model validation data
No. of patients	50	60
Sex (males/females)	36/14	40/20
Age (yr)	69.1 (12.8; 32–92)	69.2 (13.2; 41–92)
BW (kg)	57.3 (12.1; 38.4–100)	54.5 (9.73; 34.1–81)
Serum creatinine (mg/dl)	1.10 (0.86; 0.20–4.24)	1.03 (0.95; 0.26–4.13)
Creatinine clearancea (ml/min)	74.0 (54.5; 9.43–330)	85.5 (55.9; 8.88–246)
Aspartate aminotransferase (IU/liter)	47.5 (45.6; 7–226)	40.1 (35.5; 8–184)
Alanine aminotransferase (IU/liter)	43.8 (52.9; 7–293)	40.0 (58.3; 4–387)
Administration route (i.v./p.o./both)	39/8/3	56/4/0
Treatment duration (day)	11.1 (4.50; 4–23)	8.73 (3.38; 4–17)
Baseline platelet count (×103/μl)	263 (144; 71–643)c	336 (156; 127–871)
Platelet count at nadir (×103/μl)	188 (140; 26–553)c	246 (161; 39–712)
No. of patients developed thrombocytopeniab	16/45c	9/60

^aEstimated using the Cockcroft-Gault equation. i.v., intravenous infusion; p.o., oral administration (per os).

^bThrombocytopenia was defined as a decrease in the platelet count to <100 × 10³/μl.

^cFive patients were excluded from the evaluation based on the exclusion criteria.

^dData are shown as the means (standard deviations; range) unless otherwise specified.

Population pharmacokinetic model.

A total of 135 plasma concentrations from 50 patients were available to develop the PPK model. As a result of the GAM analysis, CL_CR, severe liver cirrhosis (Child Pugh grade C), sex, and age were selected for CL, and liver cirrhosis, age, and BW were selected for V as significant covariates. Finally, CL_CR and severe liver cirrhosis for CL and BW for V were significant covariates for the pharmacokinetics of LZD based on the NONMEM analysis. The regression equations for each parameter in the final model were CL (liter/h) = 2.85 × (CL_CR/60.9)^0.618 × 0.472^CIR (for CIR, 0 indicates noncirrhosis and 1 indicates cirrhosis patients) and V (liter) = 33.6 × (BW/57.9). Individual empirical Bayesian estimated CL values in the basic model (without any covariates) were significantly correlated with CL_CR (R_S [Spearman's rank correlation coefficient] = 0.690) and decreased in patients with severe liver cirrhosis (Fig. 2A and B). GOF plots show the high predictive performance of the constructed model (Fig. 3A and B), and systematic deviations were not observed (see Fig. S1 in the supplemental material). Estimated parameters are summarized in Table 2, with the median and 95% confidence interval (CI) of each parameter estimated from 1,000 bootstrap resamplings. The median values of the estimates from the bootstrapping were very similar to the population estimates in the final model, and the statistical significance of covariates was further verified by the finding that 95% CIs of all parameters did not include 1.0 or 0. Case deletion diagnostics did not identify any influential individuals for parameter estimates: all estimated parameters were within ±20% of original values.

Fig. 2.

Correlation of the individual CL of linezolid estimated by an empirical Bayesian approach with creatinine clearance (A) and severe liver cirrhosis (Child Pugh grade C) (B) in the basic model. The dashed line in panel A represents the spline curve. CIR, cirrhosis.

Fig. 3.

Goodness of fit of the PK (plasma concentration of linezolid) (A and B) and PK/PD (platelet count) (C and D) models. Population predictions were made using population mean parameters. Individual predictions were obtained using individual empirical Bayesian estimated parameters. The solid lines represents lines of identity. The dashed lines represents spline curves.

Table 2.

Estimated population pharmacokinetic/pharmacodynamic parameters for linezolid^c

Parameter	Original data			1,000 Bootstrap sample data
	Estimate (% RSE)	IIV (%; % RSE)	Shrinkagea (%)	Median	95% CI
Pharmacokinetics
ka (1iter/h)	0.583 fixedb	NE	NA	0.583 fixedb	NA
CL (liter/h)	2.85 (5.93)	35.2 (30.6)	2.89	2.85	2.55–3.26
V (liter)	33.6 (4.82)	30.8 (35.8)	30.8	33.6	30.7–37.1
PowerCLCR for CL	0.618 (15.1)	NE	NA	0.624	0.408–0.815
EffectCIR for CL	0.472 (14.2)	NE	NA	0.470	0.333–0.673
Additive error (μg/ml)	1.43 (12.5)	NE	30.5	1.42	NE
Pharmacodynamics (platelet count)
Circ0 (×103/μl)	253 (8.10)	45.9 (18.3)	0.594	254	221–298
Slope	0.00416 (16.2)	93.8 (55.8)	24.2	0.00416	0.00166–0.00601
MTT (h)	110 (9.07)	33.9 (42.8)	25.6	111	67.6–138
γ	0.203 (16.1)	NE	NA	0.212	0.0492–0.271
EffectCIR for Circ0	0.538 (12.7)	NE	NA	0.553	0.416–0.678
ShapeSlope	−0.630 (29.4)	NE	NA	−0.767	−1.63–0.180
ShapeMTT	−0.945 (59.3)	NE	NA	−1.03	−4.25–1.40
Proportional error (%)	19.8 (14.2)	NA	7.67	NE	NE
Additive error (×103/μl)	15.7 (55.9)	NA		NE	NE

^aDifference in the distribution between empirical Bayesian estimated parameter and estimated omega or sigma; refer to reference 31.

^bFixed to the value determined by Abe et al. (1).

^cRSE, relative standard error; 95% CI, 95% confidence interval; IIV, interindividual variability; Power_{CLCR for CL}, power exponent of CL_CR for CL; Effect_{CIR for CL}, fractional change of CL in severe cirrhosis (CIR) patients; Circ₀, platelet count at baseline; Slope, slope of drug effect; MTT, mean transit time; γ, feedback parameter; Effect_{CIR for Circ0}, fractional change of Circ₀ in CIR patients; Shape_Slope, shape parameter for Box-Cox transformation of slope; Shape_MTT, shape parameter for Box-Cox transformation of MTT; NE, not estimated; NA, not applicable.

Population pharmacokinetic/pharmacodynamic modeling.

A total of 603 platelet counts from 45 patients (mean, 13.4 [range, 5 to 40] counts per patient) were available for the PPK/PD modeling. Although the GAM analysis showed CL_CR, BW, and severe liver cirrhosis for Circ₀ and CL_CR for Slope to be extracted covariates, only severe liver cirrhosis for Circ₀ was selected based on the difference in OFV. GOF plots are presented in Fig. 3C and D, and they show high predictive performance and no systematic deviations (see Fig. S2 in the supplemental material). Predictive performances for individual platelet count profiles are shown in Fig. S3 in the supplemental material. Estimated parameters also are summarized in Table 2.

Model evaluation with validation data.

The actual and model-simulated platelet distribution in the validation data set (n = 60; 776 platelet counts; mean, 13.3 [range, 4 to 39] counts per patient) at each LZD level are shown in Fig. 4A. The median and variance were quite similar, indicating that the current model correctly characterized the LZD exposure-dependent reduction in platelet counts even for the external data. In addition, simulated incidences of thrombocytopenia (<100 × 10³/μl) at each treatment duration agreed well with actual incidence rates (Fig. 4B). Predicted individual nadirs also were well correlated with observed nadirs; however, in some patients, considerable overestimations were obtained (Fig. 5).

Fig. 4.

Predictive performance of the PK/PD model for external validation data. (A) Closed- and open-box whisker plots represent observed and model-predicted platelet counts versus the cumulative AUC calculated from individual estimated CL. (B) The stacked bar chart represents observed and model-predicted rates of thrombocytopenia in external validation data sorted by linezolid treatment duration. Thrombocytopenia was defined as a decrease in the platelet count to <100 × 10³/μl.

Fig. 5.

Predictive performance nadir for validation data by the PK/PD model. The solid line represents the line of identity. The dashed line represents the spline curve.

Model-based simulation for dosage adjustment.

As a result of Monte Carlo simulations based on the PPK/PD model, the probability of achieving an AUC/MIC ratio of >100 with regard to renal function (CL_CR) and dose was simulated (Table 3), and the incidence of thrombocytopenia in various conditions was predicted (Fig. 6). As shown in Table 3, 1,200 mg/day (600 mg twice daily) would be sufficient to achieve an AUC/MIC ratio of >100 in most cases, while the probability was remarkably low at 600 mg/day (300 mg twice daily) in patients with CL_CR of ≥50 ml/min. On the other hand, the risk of thrombocytopenia (<100 × 10³/μl) was remarkably high (over 30%), especially in patients with renal insufficiency (CL_CR, ≤30 ml/min) or severe liver cirrhosis (Child Pugh grade C) at 1,200 mg/day for more than 14 days after LZD treatment. These high incidences in patients with CL_CR of 30 ml/min were reduced to approximately the same level as that of patients with normal renal and hepatic functions by reducing the LZD dose to 600 mg/day, while an AUC/MIC ratio of >100 was retained. However, patients with severe insufficient renal function (CL_CR, ≤10 ml/min) or liver cirrhosis still showed a high incidence of thrombocytopenia even at 600 mg/day.

Table 3.

Simulated percentage of patients that achieved AUC/MIC ratio of >100 in each population^a (n = 1,000)

CLCR (ml/min)	CIR status	% Patients achieving AUC/MIC of >100 atdosage (mg/day):
		1,200	600
100	No	87.1	24.2
50	No	99.5	67.6
30	No	100	90.9
10	No	100	100
100	Child-Pugh grade C	99.9	90.9

^aMIC was set to 2.0 μg/ml.

Fig. 6.

Estimated incidence of thrombocytopenia each day after the start of linezolid treatment with different patient backgrounds. Thrombocytopenia was defined as a decrease in the platelet count to <100 × 10³/μl. CL_CR, creatinine clearance; CIR, severe liver cirrhosis (Child Pugh grade C).

DISCUSSION

This study shows renal function to be an important factor for LZD pharmacokinetics and a remarkable decrease in CL in patients with severe liver cirrhosis by quantitative analysis using a population modeling approach. In addition, the PPK/PD analysis quantitatively characterized the relationship between LZD exposure and LZD-induced platelet decrease and suggested dosage adjustment based on LZD levels.

Pharmacokinetic studies of LZD reported that approximately 30% of LZD is eliminated unchanged via the kidneys (1, 26, 27, 29, 33, 39), whereas the majority is metabolized by the oxidation of its morpholino ring (supposed to be mediated by ubiquitous reactive oxygen species [40]), resulting in two metabolites: an aminoethoxyacetic acid metabolite (PNU-142586) and a hydroxylethyl glycine metabolite (PNU-142300). These metabolites are excreted mainly in urine (33).

A few reports investigated the contribution of renal function to the pharmacokinetics of LZD; however, their results are controversial (7, 24, 27, 37). Brier et al. evaluated the pharmacokinetics of LZD after a single oral administration and showed that the total apparent CL of LZD did not change in patients with impaired renal function (7). On the other hand, our findings showed a significant contribution of renal function to the total CL (R_S = 0.690). Also, Meagher et al. investigated the pharmacokinetics of LZD in infectious patients with a population approach and suggested a minor contribution by renal elimination, with CL_CR explaining 16% of the total variance in LZD CL (27). In their report, LZD CL was described by a model involving two elimination pathways, parallel linear renal CL and Michaelis-Menten nonrenal CL, and they mentioned that for patients with reduced renal clearance, increased LZD concentrations caused the saturation of the nonrenal CL pathway. This might explain the discrepancy in the contribution of renal function to LZD CL among studies. In short, after a single dose, the LZD levels are not high enough to saturate the nonrenal pathway, while with repeated doses, especially in patients with insufficient renal function, accumulated LZD causes the saturation of nonrenal CL, and thus the contribution of the renal pathway is relatively great. Actually, the nonlinearity of LZD CL was reported (4, 27, 29), and elevated levels in patients with renal sufficiency were seen elsewhere, though the sample sizes were quite small (24, 37, 38). Plock et al. investigated the mechanism of the nonlinearity of LZD CL by a population PK modeling approach and suggested that this nonlinearity comes from the inhibition of its metabolism by LZD itself (29). Although their autoinhibition model was applied to our data, the GOF and OFV were not significantly improved and poor convergence performance was observed, probably because their model was too complex to apply our sparse PK data.

In this study, a remarkable decrease in CL (approximately 50% decrease) was observed in patients with severe liver cirrhosis (Child Pugh grade C). Although the number of patients with severe cirrhosis was small (n = 4), all patients showed high trough concentrations (32.5, 36.4, 40.8, and 45.4 μg/ml), suggesting that cirrhosis changes the pharmacokinetics of LZD. This finding is consistent with a report of a reduction in the CL of LZD by 60% in patients after liver transplantation/resection (35). Although the metabolism of LZD has not yet been elucidated, Wynalda et al. proposed that the major metabolite, PNU-142586, in human liver microsomes is formed through nonenzymatically (e.g., through reactive oxygen species [ROS]) and not produced by cytochrome P450 enzymes (40). It was reported that the downregulation of hepatic ROS-related protein occurs during advanced fibrosis (9). Also, liver cirrhosis is characterized by a disruption of the lobular architecture because of fibrosis, altered enzyme expression, and decreased blood flow, resulting in a reduction in the transport of drug and oxygen to hepatocytes (20). Therefore, the lower LZD CL in liver cirrhosis patients is considered to be due to these numerous pathophysiologic changes in the liver. To understand the relationship between the severity of liver cirrhosis and the pharmacokinetics of LZD, further study with a sufficient number of patients is needed.

The semimechanistic PK/PD model well characterized the exposure- and treatment duration-dependent platelet decrease even using external validation data. An earlier study showed LZD exposure- and treatment duration-dependent platelet decrease using platelet counts of 110 (13). The current results confirmed their findings using a larger data set, including external validation data. In the current model, a drug effect was defined to inhibit the proliferation of progenitor cells. So far, little is known about the mechanism responsible for LZD-induced thrombocytopenia. A few case reports suggest that LZD-induced thrombocytopenia is not responsible for bone marrow suppression based on the observation that the patients with thrombocytopenia retained an adequate number of megakaryocytes in their bone marrow (5, 11). Bernstein et al. speculated that LZD-induced thrombocytopenia is caused by a quinine/quinidine mechanism of immune-mediated platelet destruction based on an improved rate of decline in the platelet count following immunoglobulin (IVIG) therapy (5).

The estimated parameters representing the system-dependent component (i.e., MTT and γ) were similar to those reported previously (MTT = 142 h, γ = 0.176) (41). The 95% CI of the parameter representing drug effect (Slope) did not include 0, suggesting that the exposure level is related to the platelet decrease. Meanwhile, interindividual variability in Slope was large (∼100%), and individual predictions for the nadir gave considerable overestimations in several cases, leading to an underestimation of the risk of thrombocytopenia. These results suggest that platelet profiles are not determined only by exposure to LZD, and there is substantial variability in individual sensitivity to LZD. Thus, the careful monitoring of platelet counts still is needed for the safe use of LZD.

Simulation based on the PK/PD model revealed 1,200 mg/day (600 mg twice daily; current standard dosage) to be appropriate for patients with sufficient renal function (CL_CR of ≥50 ml/min) and without cirrhosis; however, a remarkably increased risk was found for thrombocytopenia in patients with insufficient renal function (CL_CR of ≤30 ml/min) or severe liver cirrhosis (Child-Pugh grade C). The estimated probability of thrombocytopenia at 14 days after the start of LZD treatment (1,200 mg/day) in patients with insufficient renal function (CL_CR = 30 to 10 ml/min) was 32.6 to 51.0%, which is comparable with the incidence reported previously (64.7% in insufficient renal function at 17.4 ± 10.3 days) (10). For these high-risk patients, the current standard dose of 1,200 mg/day is too high considering its safety, and a reduced regimen of 600 mg/day (300 mg twice daily) would be appropriate for efficacy based upon the AUC/MIC ratio. However, in patients with severe renal insufficiency (CL_CR of ≤10 ml/min) or severe liver cirrhosis, high incidences of thrombocytopenia were estimated even at 600 mg/day due to elevated exposure levels and low baseline platelet counts, suggesting that the careful monitoring of platelet counts is strongly recommended for those patients.

In summary, the population pharmacokinetic analysis revealed that renal function and severe liver cirrhosis significantly affect the pharmacokinetics of LZD. The population PK/PD analysis successfully and quantitatively described LZD exposure-dependent platelet decreases. Furthermore, a simulation study based on a newly constructed model suggested a reduced dose (600 mg/day) for patients with insufficient renal function (CL_CR of ≤30 ml/min) or severe liver cirrhosis (Child-Pugh grade C). The present study demonstrated the need for the dosage adjustment of LZD based on population PK/PD modeling.