Population pharmacokinetics, safety and dosing optimization of voriconazole in patients with liver dysfunction: a prospective study

Aims This study aimed to explore the relationship between voriconazole trough concentration (Ctrough) and toxicity, identify the factors signiﬁcantly associated with voriconazole pharmacokinetic parameters and propose an optimised dosing regimen for patients with liver dysfunction. Methods The study prospectively enrolled 51 patients with 272 voriconazole concentrations. Receiver operating characteristic (ROC) curves were used to explore the relationship between voriconazole Ctrough and toxicity. The pharmacokinetic data was analysed with nonlinear mixed-eﬀects method. Dosing simulations stratiﬁed by TBIL (TBIL-1: TBIL < 51 μ mol/L; TBIL-2: 51 μ mol/L [?] TBIL < 171 μ mol/L; TBIL-3: TBIL [?] 171 μ mol/L) were performed. Results ROC curve analysis revealed that voriconazole Ctrough of [?] 5.1 mg/L were associated with signiﬁcantly lower the incidence of adverse events. A one-compartment pharmacokinetic model with ﬁrst-order absorption and elimination was used to describe the data. Population pharmacokinetic parameters of clearance (CL), the volume of distribution (V) and oral bioavailability (F) were 0.88 L/h, 148.8 L and 88.4%, respectively. Voriconazole CL was signiﬁcantly associated with total bilirubin (TBIL) and platelet count. The V increased with weight. Patients with TBIL-1 could be treated with loading dose of 400 mg every 12 hours (q12h) for ﬁrst day and maintenance dose of 100 mg q12h intravenously or orally. TBIL-2 and TBIL-3 patients could be treated with loading dose of 200 mg q12h and maintenance doses of 50 mg q12h or 100 mg once daily (qd) and 50 mg qd orally or intravenously, respectively. Conclusions TBIL-based dosing regimens provide a practical strategy for voriconazole maximizing treatment outcomes.


Introduction
Infections are common and represent one of the most important reasons of progression of liver failure, development of liver-related complications, and mortality in patients with liver dysfunction [1]. Invasive fungal infections can be a life-threatening complication in patients with liver dysfunction and are associated with a high morbidity and significant mortality [2][3][4][5]. Furthermore, long-term use of broad-spectrum antibiotics and glucocorticoids, invasive procedures including liver puncture, ascites drainage, indwelling catheters and hemodialysis, and multiple hospitalizations are also associated with an increased risk of invasive fungal infections [6] and are common in patients with liver dysfunction.
Voriconazole is a triazole antifungal agent that exhibits broad-spectrum activity and is used for both the prevention and treatment of invasive fungal infections [7]. Metabolism of voriconazole occurs in the liver by hepatic cytochrome P450 isoenzymes, primarily CYP2C19 and to a lesser extent CYP3A4 and CYP2C9 [8]. Multiple factors are already known to be associated with variability in voriconazole pharmacokinetics, including age, weight, liver function and genetic polymorphism of the CYP2C19 enzyme [9][10][11][12]. Voriconazole exhibits complex nonlinear pharmacokinetics and has a narrow therapeutic window [13,14]. Subtherapeutic concentrations have been associated with therapeutic failure, and supratherapeutic concentrations are correlated with an increased risk of neurological, visual and hepatic toxicity [14,15]. Therapeutic drug monitoring (TDM) of voriconazole is advocated to improve treatment outcomes and minimize the risk of adverse events. As the liver plays a key role in the disposition of voriconazole including absorption, distribution, metabolism and excretion [16], liver dysfunction can change the pharmacokinetic characteristics of voriconazole, increasing the risk of voriconazole accumulation and subsequent adverse events.
The voriconazole product information suggests that patients with mild-to-moderate liver dysfunction (Child-Pugh class A and B) should receive half of the maintenance dose after an unchanged loading dose. However, there is limited information about the pharmacokinetics and appropriate dosing of voriconazole in patients with severe liver dysfunction (Child-Pugh class C). We have previously demonstrated that the clearance of voriconazole was significantly decreased in patients with liver dysfunction ]17] highlighting the necessity to optimise voriconazole dosing regimens in these patients.
Population pharmacokinetic (PPK) analysis was used to evaluate the pharmacokinetic characteristics and identify the measurable factors of patient-related and clinical-related pharmacokinetic variabilities. Monte Carlo simulation (MCS) is a valuable tool to determine dosing regimens and optimize antibacterial therapies [18]. The present study aims to: 1) develop a PPK model of voriconazole in patients with liver dysfunction; 2) identify factors significantly associated with voriconazole pharmacokinetic parameters; 3) explore the relationship between voriconazole trough concentration (C trough ) and toxicity to identify the safety C trough range; 4) evaluate potential voriconazole dosing regimens in patients with liver dysfunction through Monte Carlo Simulation (MCS) utilizing final pharmacokinetic model.

Patients and Data collection
The prospective and observational study was conducted on liver dysfunction patients who received voriconazole between February 28, 2018 and December 11, 2018. The inclusion criteria were (1) Age[?]15 years; (2) Patients were diagnosed with liver dysfunction, such as liver failure or liver cirrhosis according to the Child-Pugh classifications; (3) Treatment or prevention of invasive fungal infections with voriconazole; (4) patients contributed at least one blood sample. The exclusion criteria were: (1) Patients who were allergic or intolerant to voriconazole; (2) Pregnant or lactating patients; (3) Using potent CYP450 inducer or inhibitor such as rifampicin, isoniazid, phenytoin, carbamazepine during voriconazole treatment, but did not include proton pump inhibitors (PPIs). (4) Patients who lacked the necessary data such as genotype of CYP2C19, renal and liver function index. This study was approved by Ethics Committee of The Second XiangYa Hospital of Central South University (Changsha, China). All of the patients were provided written informed consent before participating in the study.

Dosing regimen and Specimen collection
Voriconazole dosing was according to the product information, where patients with mild to moderate liver dysfunction (Child-Pugh A and B) received standard loading doses (400 mg twice daily PO or 6 mg/kg twice daily IV) on the first day, but half the standard maintenance dose (100 mg twice daily PO or 2 mg/kg twice daily IV). Due to the limited data on the dosing of voriconazole in patients with severe liver dysfunction (Child-Pugh C), dosing of these patients was based on the clinician's experience. The subsequent doses for all patients were adjusted according to the measured voriconazole trough concentration (C trough ) and the patient's clinical response to voriconazole (effective or ineffective, with or without adverse effects).
Venous blood samples (2 mL) were collected into anticoagulant tubes. Patients were randomly collected 2-3 blood samples without intervention in treatment at 0.5 h, 1 h, 1.5 h, 2 h, 4 h, 6 h, 8 h, 12 h, 24 h after intravenous or oral administration. In addition, blood samples such as C trough from TDM were collected from all patients after 24 hours. All voriconazole plasma concentrations were analysed by automatic twodimensional liquid chromatography (2D-HPLC, Demeter Instrument Co., Ltd., Hunan, China) as previously described [17].

Statistical analysis
The Wilcoxon two-sample test or Kruskal-Wallis test was used to compare voriconazole C trough . Proportions were compared with the Chi-square test or Fisher's exact test. Univariate analysis was performed to assess the association between voriconazole C trough and adverse events. Receiver operating characteristic (ROC) curves were used to explore the relationship between voriconazole C trough and adverse events. Statistical analysis was performed with SPSS version 22.0 (IBM Corporation, Armonk, New York).

Population Pharmacokinetics analysis
The concentration-time data of voriconazole was developed using Phoenix NLME (version 8.0, Pharsight Corporation, USA). The first-order conditional estimation method with the η-ε interaction option (FOCE ELS) was used throughout the model development.
One-and two-compartment structural kinetic models with first-order and Michaelis-Menten elimination were evaluated to describe the pharmacokinetics of voriconazole. Finally, we comprehensively compared the objective function value (OFV), graphical goodness of fit, the evaluation of parameter estimates (including precision) and scientific and physiological plausibility to choose the best base model. The oral absorption rate constant (k a ) was fixed to a value of 1.1 h -1 based on the results from a previous study [23].
The inter-individual variability in voriconazole pharmacokinetic parameters was described with an exponential error model. Residual error models for voriconazole were tested as follows: the proportional error model, the additive error model and combined error model, including proportional plus additive error model. Potential demographic and biochemical covariates were evaluated by visual inspection of covariates possible relationships with pharmacokinetic parameters included in the model. For continuous covariates, a linear, piece-wise, exponential, and power parameter-covariate relations were tested. Categorical covariates were linearly included. Then, a covariate model in a stepwise forward-inclusion and backward-elimination procedure were carried out. A covariate was considered to be significant when inclusion of the covariate resulted in a decrease in the objective function value (OFV) of greater than 6.64 (p< 0.01) and elimination of the covariate resulted in an increase in the OFV of greater than 10.83 (p< 0.001).
Goodness-of-fit (GOF) plots were used to evaluate the adequacy of fitting. The bootstrap method was used to assess the robustness and stability of the final model. 1000 resamples from the original data were performed. All of the model parameters were estimated, and their median and 2.5 and 97.5 percentiles were calculated. That was stable if the 95% CI for the parameter estimates derived from the 1000 bootstrap runs encompassed the original final parameter estimate.

Results
Patients' characteristics 51 patients with a total of 272 voriconazole plasma concentrations were included in this study. The demographics and clinical information of the patients is summarized in Table 1 . Patients with Child-Pugh grade C or MELD score greater than 15 scored made up more than 70% of all patients. There was a significant variation in the voriconazole plasma concentrations, with an average concentration of 3.9 mg/L and a range of 0.06-14.08 mg/L. There were 190 plasma C trough , and 82 plasma concentrations collected within the 24 hours after intravenous or oral administration. There were four types of CYP2C19 genotypes in the present study, 1 UM patients (CYP2C19*17*17), 24 EM patients (CYP2C19*1*1), 21 IM patients (CYP2C19*1*2, CYP2C19*1*3), and 5 PM patients (CYP2C19*2*2, CYP2C19*2*3). The genotypes were divided into three groups (UM/EM, IM and PM) for the purposes of PPK model development.

Voriconazole concentrations and adverse events
Adverse events were reported in 20 patients (39.2%) during voriconazole therapy. These included dizziness, hallucinations and visual disturbance such as altered colour discrimination, blurred vision and photophobia.
The median duration from voriconazole initiation to onset of adverse events was 2 days (range, 1 to 12 days). The median voriconazole concentration at the time of these adverse events was significantly higher than in patients without adverse events (6.5 mg/L versus 2.3 mg/L, P <0.0001). A ROC curve analysis confirmed voriconazole C trough to be a significant predictor of adverse events, with a voriconazole C trough of [?] 5.1 mg/L found to minimize the incidence of adverse events (Figure 1).

Population Pharmacokinetic Analysis
A one-compartment pharmacokinetic model with first-order oral absorption and elimination adequately describe the data. Inter-individual variability of the parameters was best fitted to an exponential equation, and residual error was best characterized by a proportional error model.  Table 2.
Goodness-of-fit plots from the basic and final models presenting the correlations between populationpredicted concentrations and individual-predicted versus observed concentrations of voriconazole are showed in Figure 2. The figure shows improvement in the final model fit has been improved compared to the base model. There was no structural bias in the plot of population-predicted and individual-predicted concentrations versus observed concentrations. The conditional weighted residuals (CWRES) of population-predicted concentrations and time for voriconazole are showed in Figure 3. The CWRES random distribution was around zero for voriconazole. The distribution was symmetrical distribution and no concentration-or timerelated trends were observed for voriconazole. Most of points were within an acceptable range (-2 to 2).
The bootstrap (n=1000) procedure is summarized in Table 2

Monte Carlo simulation
The elimination of voriconazole is markedly prolonged (typical value of CL: 0.88 L/h) in patients with liver dysfunction, which means it reaches the steady state about 30 days later. Furthermore, fungal infection treatment usually takes one month or more. Therefore, the dosing regimens were simulated at 30-days for treatment. Simulations of oral or intravenous administration did not demonstrated a significant difference. The probability of C trough target attainment after intravenous and oral administration for 30 days of standard unadjusted dosing regimen of voriconazole for patients without liver dysfunction (Loading dose: 400 mg q12h, maintenance dose: 200 mg q12h) are showed in Table 3. The maximum PTA of all group is less than 50%. Apart from TBIL-1 patients, there was 90% overexposure in the other groups. The results for the recommended dosing regimen of voriconazole for patients with mild to moderate liver dysfunction (Child-Pugh A and B) (Loading dose: 400 mg q12h, maintenance dose: 100 mg q12h) are showed in Table 4. The PTA for patients with TBIL-1 is 91.7% and 85.2%, administered orally and intravenously respectively. It indicated that dosing regimen with a loading dose of 400 mg q12h for 2 doses, followed by a maintenance dose of 100 mg q12h administered intravenously or orally for patients with TBIL-1 is suitable.
For patients with TBIL-2 and TBIL-3, we simulated the achievement of C trough after oral and intravenous administration with different loading doses (400 mg, 300 mg and 200 mg q12h) in order to determine the loading dose, which are showed in Table 5. An oral and intravenous loading dose of 200 mg q12h demonstrated the highest PTA (¿90%). Utilizing this loading dose, different maintenance doses and dosing intervals were simulated to determine the optimal maintenance dose. The PTA of the examined maintenance doses are showed in Table 6. The simulations demonstrated that a maintenance dose of 50 mg q12h or 100 mg qd orally or intravenously for TBIL-2 patients, and a maintenance dose of 50 mg qd orally or intravenously for TBIL-3 patients were optimal. The simulated 30 days median voriconazole C trough versus time profiles based on the optimal intravenous or oral dosing regimen are showed in Figure 4. The results showed that the median voriconazole C trough in all patients were within the target concentration range (0.5-5.0 mg/L), and the distribution of C trough was centralized between 2 and 4 mg/L.

Discussion
This study prospectively investigated 51 patients with liver dysfunction to develop a PPK model. The results of this analysis show that a one-compartment pharmacokinetic model with first-order absorption and elimination was able to describe voriconazole pharmacokinetics in patients with liver dysfunction.
The estimated values of the pharmacokinetic parameters CL, V, and F of voriconazole in patients with liver dysfunction (0.88 L/h, 148.8 L and 88.4%, respectively) are similar to our previous findings [17] (0.56 L/h, 134 L and 80.8%, respectively). We confirmed that voriconazole shows a significant decrease in CL in patients with liver dysfunction compared with patients without liver disease and healthy subjects (CL: 4.76-25.2 L/h) [10,13,25,26]. The V is not significantly different in the presence of liver disease. Covariate model showed that TBIL and PLT are significantly associated with voriconazole CL, while WT has a significant effect on V. The inclusion of TBIL, PLT and WT reduced the inter-individual variation of CL and V (the interindividual variation of CL decreased from 68.3% to 18.0%, and the inter-individual variation of V decreased from 15.3% to 12.0%), indicating that these covariates are important factors affecting the large variation of pharmacokinetic parameters.
TBIL was showed to be an important covariate affecting the CL of voriconazole in this study, the inclusion of TBIL resulted in a significant reduction of the OFV (ΔOFV=71.36) in the forward inclusion modelbuilding step. The final model demonstrated that high TBIL values were significantly correlated with decreased CL. Voriconazole is mainly metabolized by the CYP450 enzyme in the liver (98%) and then excreted through the kidney and bile, with less than 2% of a dose of voriconazole is excreted into the urine as unchanged voriconazole [27]. In liver disease, a reduction in absolute liver cell mass or a decreased in metabolic enzyme activity may lead to impaired drug metabolism [16], which causes a large amount of voriconazole to accumulate in the body. Therefore, voriconazole CL is significant decrease for patients with liver dysfunction. The PLT was found to be significantly associated with CL in the present study, similar to our previous studies [17]. The reduction of PLT counts is very common in patients with cirrhosis and is correlated with severity of liver function. WT has a significant effect on V, and is positively correlated with V.
Age, CYP2C19 genotype, and PPI were not found to affect significantly the pharmacokinetic parameters of voriconazole, which is consistent with our previous analysis 17 . A prospective study of voriconazole by Wang et al. [28] has shown that age has a significant effect on voriconazole CL, the median voriconazole plasma concentrations in elderly (age [?]65 years) have been 80%-90% higher than those in younger patients. Another prospective study of lung transplant recipients [29] found a correlation between age and initial voriconazole C trough , the older patients (age [?] 60 years) is more likely to have a higher initial C trough . In older patients, the activity of liver microsomal enzyme is decreased, resulting in lower CL. However, this study did not find age to have a significant effect on the pharmacokinetic parameters of voriconazole, suggesting that age has no significant effect on liver microsomal enzymes in patients with liver dysfunction. Many studies [30][31][32][33] in patients without liver disease have showed that PM patients have higher voriconazole plasma concentration compared with EM and IM patients. However, CYP2C19 polymorphisms and PPI (CYP2C19 enzyme inhibitors) seem to have no effect on the pharmacokinetic parameters of voriconazole in this study. Ohnishi et al. [34] have reported that in 31 patients with chronic liver disease (9 with chronic hepatitis, 22 with cirrhosis comprising 20 Child-Pugh type A, 1 type B, 1 type C), patients with PM polymorphisms have higher omeprazole hydroxylation indexes (a metabolite of CYP2C19 enzyme) than those with EM and IM polymorphisms, but only two Child-Pugh B and C patients were included. In patients with moderate to severe liver dysfunction, whether gene polymorphism is still an important factor affecting CYP2C19 enzyme activity is worthy of further investigation.
At present, the product information for voriconazole suggests that the standard loading dose should be used but the maintenance dosing should be halved in patients with mild-to-moderate liver disease (Child-Pugh Class A and B), however no dose recommendations in severe liver dysfunction patients are provided. It has been reported in a retrospective study [35] that oral voriconazole maintenance doses in patients with Child-Pugh class C should be reduced to approximately one-third that of patients with normal liver function, while another clinical study for acute-on-chronic liver failure (ACLF) patients [4] has proposed that voriconazole concentration can be maintained a reasonable range (1-5 mg/L) with a loading dose of 200 mg twice daily and a maintenance dose of 100 mg once daily of voriconazole dosing regimen. However, both of these studies are retrospective analyses with small sample sizes (6 cases of cirrhosis C grade and 20 cases of chronic acute liver failure, respectively), so the voriconazole dosing regimen for patients with liver dysfunction still needs further verification.
In the current study, TBIL-based simulations after intravenous and oral voriconazole were performed using voriconazole C trough (0.5-5.0 mg/L) as a target with the combination of MCS to optimize voriconazole dosing regimen. The results show that there is no significant difference in the PTA after voriconazole intravenous and oral administration. The dosing regimen for patients with normal liver function (loading dose: 400 mg q12h; maintenance dose: 200 mg q12h) is probably inappropriate for patients with liver dysfunction, and is associated with a high risk of toxicity (51.6%-97.8% probability of toxicity). Patients with TBIL-1 could be treated with loading dose of 400 mg q12h for 2 doses followed by maintenance dose of 100 mg q12h intravenously or orally which is the dosing regimen of patients with mild-to-moderate liver disease (Child-Pugh Class A and B) in the medication label of voriconazole, but it's not suitable for patients with TBIL-2 and TBIL-3. For patients with TBIL-2 and TBIL-3, the PTA of voriconazole within 30 days is greater than 90% when TBIL-2 and TBIL-3 patients could be treated with maintenance doses of 50 mg q12h or 100mg qd and 50 mg qd orally or intravenous, respectively. Meanwhile, the steady-state time (about 30 days) of voriconazole is markedly prolonged in patients with liver dysfunction, a loading dose of 200 mg q12h orally or intravenously must be given to rapidly achieve the voriconazole target concentration.
This study found that adverse events have generally occurred at higher voriconazole concentrations, and ROC curve analysis revealed a significant association between voriconazole C trough and toxicity, with voriconazole C trough of [?] 5.1 mg/L found to minimize the incidence of adverse events, similar to the studies by Dolton et al [36]. and Troke et al [37].
There are several limitations to the present study. Firstly, this study has a small sample size and it is a single-center study. Secondly, this study did not find the CYP2C19 genotype to have a significant effect on the pharmacokinetic parameters of voriconazole, possibly due to the small number of patients with PM and UM polymorphisms included. Thus, the results need further validation in future clinical studies.

Conclusions
This study suggests that the TBIL, PLT and WT are significantly associated with voriconazole pharmacokinetic parameters. TBIL is a critical factor leading to large pharmacokinetic variation of voriconazole. Using MCS to optimize the dosing regimen in patients with liver dysfunction based on our PPK model and TBIL stratification we demonstrated that lower doses and longer administration intervals should be considered for patients with liver dysfunction. This is helpful for clinicians making decisions about voriconazole dosing regimens, especially to determine efficient initial dosing strategies and in primary hospitals where TDM is not available.