Model-Oriented Dose Optimization of Voriconazole in Critically Ill Children

ABSTRACT

This study aimed to employ a population pharmacokinetic (PK) model to optimize the dosing regimen of voriconazole (VRC) in children with a critical illness. A total of 99 children aged from 0.44 to 13.58 years were included in this study. The stability and predictive performance of the final model were evaluated by statistical and graphical methods. The optimal dosing regimen was proposed for children with different body weights, CYP2C19 phenotypes, and coadministrations with omeprazole. The PK of VRC was described by a two-compartment model with nonlinear Michaelis-Menten elimination. Body weight, CYP2C19 phenotype, and omeprazole were significant covariates on the maximum velocity of elimination (V_max), which had an estimated typical value of 18.13 mg · h⁻¹. Bayesian estimation suggested that the dose-normalized concentration and total exposure (peak concentration [C_max]/D, trough concentration [C_min]/D, and area under the concentration-time curve over 24 h [AUC₂₄]/D) were significantly different between extensive metabolizer (EM) patients and poor metabolizer (PM) patients. To achieve the target concentration early, two loading doses of 9 mg · kg⁻¹ of body weight every 12 h (q12h) were reliable for most children, whereas three loading doses of 6 to 7.5 mg · kg⁻¹ q8h were warranted for young children weighing ≤18 kg (except for PM patients). The maintenance doses decreased about 30 to 40% in PM patients compared to that in EM patients. For children aged <2 years, in EM patients, the maintenance dose could be as high as 9 mg · kg⁻¹. The maintenance dose of VRC was supposed to decrease slightly when coadministered with omeprazole. A population PK model of intravenous VRC for critically ill children has been successfully developed. It is necessary to adjust dosing regimens according to the CYP2C19 genotype. Optimal dosing regimens have been recommended based on the final model.

TEXT

Invasive fungal disease (IFD) is a serious life-threatening infectious disease. It leads to high morbidity and mortality rates in pediatric patients who have a compromised immune system (1, 2). Due to unavoidable factors such as chemotherapy, the presence of central venous catheters, the use of high-dose corticosteroids, and broad-spectrum antibiotic therapy, patients with hematological malignancies and pediatric intensive care unit patients have a particularly high risk of IFD (3). Voriconazole (VRC), a broad-spectrum antifungal agent used for the treatment and prophylaxis of IFDs, is recommended as the first-choice drug for the treatment of invasive aspergillosis and an alternative drug for the treatment of candidemia (4).

However, VRC has a narrow therapeutic range. Several studies demonstrated that a trough concentration (C_min) of <1.0 μg · ml⁻¹ will reduce the rate of success, while a C_min of >5.5 μg · ml⁻¹ will increase the rate of toxicity (1, 4, 5). Apart from the fact that over 60% of pediatric patients have subtherapeutic exposures, the observed intra- and interindividual variabilities are larger in children than in adults (1, 6, 7). These variabilities are attributed not only to physiological factors such as age, body weight, CYP2C19 genotype, liver function, and maturation of organ function but also to commonly seen disease factors in patients with critical illness, including drug interactions, inflammation, and hypoproteinemia.

Therapeutic drug monitoring (TDM) is a suitable approach to correct inadequate exposures that may result in reduced clinical efficacy and to avoid adverse reactions. But continuous sampling and the time delays are disadvantages of TDM and should not be ignored. Population pharmacokinetic (PK) studies can provide individual PK parameters, which are helpful to predict individualized doses and maintain drug exposure at the expected level. Moreover, critically ill children can benefit from the avoidance of multiple samplings.

Previous PK studies have focused on children aged 2 to 12 years and teenagers weighing <50 kg (4, 6). The previously recommended dose was a loading dose of 9 mg · kg⁻¹ of body weight intravenously (i.v.) every 12 h (q12h) for the first 24 h, followed by a maintenance dose of 8 mg · kg⁻¹ i.v. q12h, which is also recommended to be 7 mg · kg⁻¹ i.v. q12h by some researchers (1, 8). However, limited data are available regarding optimal VRC dosing and PK studies in Chinese pediatric patients. Due to the ethnic differences, we cannot assume that it is easy to transfer the previously reported PK models to Chinese children. Here, we performed a population PK analysis by recruiting a large number of severely ill children in China, aiming to identify the factors that influence the PK variability of VRC and optimize the current dosing regimens in this special population.

RESULTS

A total of 195 serum concentrations were obtained from 99 children with a median age of 5.25 years (range, 0.44 to 13.58 years). A summary of patient demographics and clinical characteristics is presented in Table 1. All enrolled patients received intravenous VRC at a median loading dose of 6 mg · kg⁻¹ (range, 3.6 to 9.8 mg · kg⁻¹) q12h on day 1, followed by a median maintenance dose of 4 mg · kg⁻¹ (range, 3.6 to 9.0 mg · kg⁻¹) q12h. Of 73 serum trough concentrations detected, only 24 (32.9%) concentrations achieved the target level, and 42 (57.5%) were at subtherapeutic levels. Among all the subjects, only one patient (CYP2C19*1/*17) belonged to the ultrarapid metabolizer (UM) group, one patient carried an infrequent allele of CYP2C19*2/*17, and four children had not done a genotyping test. Therefore, all six of these patients were classified as an independent group and not investigated in group comparisons. In addition, deaths occurred in two patients. These happened during VRC treatment and within 7 days after withdrawing VRC, respectively. Visual disturbance was observed in one child.

TABLE 1.

Demographic and clinical characteristics of patients (n = 99)

Parameter	Value
	Mean (SD)	Median (range)
No. of samples	195
No. of male patients	60
Age (yrs)	6.14 (3.48)	5.25 (0.44–13.58)
Age <2 yrs (n = 8)	1.06 (0.54)	0.92 (0.44–1.92)
WT (kg)	19.9 (8.4)	18.0 (7.0–41.0)
ht (cm)	113.5 (22.7)	112.0 (66.0–159.0)
Body surface area (m2)	0.79 (0.24)	0.74 (0.36–1.31)
Aspartate aminotransferase level (U · liter−1)	50.4 (78.9)	28.0 (2.0–590.0)
Alanine aminotransferase level (U · liter−1)	55.6 (119.1)	19.0 (1.0–926.0)
Total bilirubin level (μmol · liter−1)	12.1 (10.2)	8.5 (2.7–61.3)
Albumin level (g · liter−1)	35.6 (5.0)	35.9 (23.8–51.3)
Blood urea nitrogen level (mmol · liter−1)	4.4 (2.0)	4.10 (0.7–12.2)
Serum creatinine concn (μmol · liter−1)	25.7 (9.0)	22.9 (9.0–67.1)
Uric acid level (μmol · liter−1)	189.2 (97.8)	160.0 (35.1–585.0)
Estimated glomerular filtration rate (ml · min−1 · 1.73 m−2)	177.6 (58.8)	172.9 (61.4–393.5)
C-reactive protein level (mg · liter−1)	43.3 (55.9)	20.4 (0.8–276.0)
Interleukin-6 level (pg · ml−1)	157.7 (673.5)	30.2 (3.5–6,822.2)
VRC treatment duration (days)	10.8 (5.3)	9.5 (3.5–30.0)
VRC loading dose (mg · kg−1)	6.6 (1.4)	6.0 (3.6–9.8)
VRC maintenance dose (mg · kg−1)	5.2 (1.7)	4.0 (3.6–9.0)
VRC concn (mg · liter−1)	1.79 (2.24)	0.88 (0.08–12.22)
No. of patients with underlying disease
Acute lymphoblastic leukemia	36
Acute myeloblastic leukemia	20
Aplastic anemia	9
Monocytic leukemia	6
Other	28
No. of patients with CYP2C19 phenotype
Ultrarapid metabolizer	1
Extensive metabolizer	34
Intermediate metabolizer	45
Poor metabolizer	14
No. of samples with concomitant medication
Omeprazole	38
Glucocorticoids	68

Population PK model development.

A two-compartment model with nonlinear Michaelis-Menten elimination could better describe the PK data. The model was parameterized in terms of the central volume of distribution (V₁), peripheral volume of distribution (V₂), Michaelis constant (K_m), intercompartment clearance (Q), and maximum velocity of elimination (V_max). K_m was fixed to 1.15 μg · ml⁻¹ based on previous research (9, 10).

Considering the influence of developmental changes, six allometric models were evaluated by investigating the relationship between V_max and weight (WT) (or age). As shown in Table 2, the simple WT-based exponent model (model II) worked best, which had the lowest objective function value (OFV), Akaike information criterion (AIC) value, and Bayesian information criterion (BIC) value. After the incorporation of covariates, WT, CYP2C19 phenotype, and omeprazole were identified as determinant variables for V_max, guided by significant drops in the OFV of 46.19, 28.52, and 7.91 points, respectively. In the final model, the interindividual variability was 8.19% for V_max, and the residual variability was 58.20%. No time-dependent relationship was confirmed for V_max by testing model VII, which led to a higher OFV (464.17) than that obtained with the final model (OFV, 459.76).

TABLE 2.

Six candidate models for V_max and their parameter estimates^a

Parameter	Value
	Model I: 3/4 allometric model	Model II: simple WT-based exponent model	Model III: simplest BSA-based exponent model	Model IV: maturation model	Model V: WT-dependent exponent model	Model VI: age-dependent exponent model
Model description	$V_{\text{max}}={\text{θ}}_{V_{\text{max}}}\times {\left(\frac{\text{WT}}{{\text{WT}}_{\text{median}}}\right)}^{k_1}\times \text{MF}$
k 1	0.75	Estimated	Estimated	0.75	$k_0-\frac{k_{\text{max}}\cdot {\text{WT}}^{\text{γ}}}{k_{50}^{\text{γ}}+{\text{WT}}^{\text{γ}}}$	$k_0-\frac{k_{\text{max}}\cdot {\text{age}}^{\text{γ}}}{k_{50}^{\text{γ}}+{\text{age}}^{\text{γ}}}$
MF	1	1	1	$\frac{1}{1+{\left(\frac{\text{age}}{{\text{TM}}_{50}}\right)}^{-\text{γ}}}$	1	1
OFV	547.02	538.09	538.40	544.62	540.82	543.22
AIC	565.02	558.09	558.40	566.62	566.82	569.22
BIC	594.48	590.82	591.13	602.62	609.37	611.77
${\text{θ}}_{V_{\text{max}}}$ (% SE)	13.30 (13.51)	13.31 (6.44)	13.43 (7.56)	18.86 (10.27)	14.89 (10.89)	14.48 (43.13)
TM50 (% SE)	NA	NA	NA	3.52 (21.45)	NA	NA
γ (% SE)	NA	NA	NA	0.89 (17.56)	2.20 (6.37)	6.46 (12.93)
k0 (% SE)	NA	NA	NA	NA	1.29 (37.61)	1.09 (17.25)
kmax (% SE)	NA	NA	NA	NA	1.25 (23.98)	0.73 (8.69)
k50 (% SE)	NA	NA	NA	NA	16.50 (13.72)	2.84 (8.51)

^a${\text{θ}}_{V_{\text{max}}}$, typical value of the maximum velocity of elimination (V_max); k₁, allometric exponent; MF, factor for maturation; TM₅₀, maturation half-time; γ, Hill coefficient defining the steepness of the sigmoidal curve; k₀, the exponent at a theoretical weight of 0 kg or age of 0 years; k_max, maximum decrease of the exponent; k₅₀, the weight or age achieving a 50% drop in the maximum decrease of the exponent. NA, not applicable.

The final model is represented as follows:

$$V_1={\text{θ}}_{V_1}\times {\left(\frac{\text{WT}}{18}\right)}^{{\text{θ}}_1}$$ (1)

$$V_2={\text{θ}}_{V_2}\times {\left(\frac{\text{WT}}{18}\right)}^{{\text{θ}}_2}$$ (2)

$$Q={\text{θ}}_Q\times {\left(\frac{\text{WT}}{18}\right)}^{{\text{θ}}_3}$$ (3)

$$V_{\text{max}}={\text{θ}}_{V_{\text{max}}}\times {\left(\frac{\text{WT}}{18}\right)}^{{\text{θ}}_4}\times \text{exp}\left({\text{θ}}_{\text{CYP}}\right)\times \text{exp}\left({\text{θ}}_{\text{inh}}\right)$$ (4)

where V₁, V₂, Q, and V_max represent the PK parameters and WT is body weight in kilograms. θ_CYP and θ_inh are estimated parameters describing the fixed effect of the CYP2C19 phenotype and enzyme inhibitor on V_max. Table 3 exhibits the PK parameters estimated from the final model. The population-typical values of V₁, V₂, Q, and V_max were 22.79 liters, 61.28 liters, 13.71 liters · h^-1, and 18.13 mg · h⁻¹, respectively. As shown in Fig. 1, significant differences in weight-normalized V_max were obtained among three subpopulations grouped by CYP2C19 phenotype (extensive metabolizer [EM], intermediate metabolizer [IM], and poor metabolizer [PM]), with P < 0.001 (F = 13.74).

TABLE 3.

Population PK parameters of VRC and bootstrap results^a

Parameter
	Final model		Bootstrap analysis
	Estimate	SE (%)	2.5%	Median	97.5%	Bias (%)
${\text{θ}}_{V_1}$ (liters)	22.79	38.99	5.24	22.14	40.96	−2.85
${\text{θ}}_{V_2}$ (liters)	61.28	31.96	22.82	62.30	103.58	1.66
${\text{θ}}_{K_m}$	1.15 (fixed)			1.15 (fixed)
θQ (liters/h)	13.71	40.89	2.71	13.73	24.74	0.15
${\text{θ}}_{V_{\max }}$ (mg/h)	18.13	7.75	14.84	17.73	20.76	−2.21
θ1	1 (fixed)			1 (fixed)
θ2	1 (fixed)			1 (fixed)
θ3	0.75 (fixed)			0.75 (fixed)
θ4	0.76	14.52	0.44	0.77	1.08	1.32
θCYP-EM	0 (fixed)			0 (fixed)
θCYP-IM	−0.12	8.57	−0.16	−0.12	−0.08	0
θCYP-PM	−0.38	41.62	−0.74	−0.39	−0.04	2.63
θinh-NON	0 (fixed)			0 (fixed)
θinh	−0.21	42.35	−0.42	−0.22	−0.03	4.76
Interindividual variability (%)
${\text{ω}}_{V_1}$	7.74	7.23	6.52	8.27	10.09	6.85
${\text{ω}}_{V_2}$	55.59	18.85	35.05	59.86	84.45	7.68
ωQ	160.30	11.77	119.65	159.71	192.95	−0.37
${\text{ω}}_{V_{\text{max}}}$	8.19	31.45	2.94	9.23	15.68	12.70
Residual variability (%)
σ	58.20	13.25	42.14	58.29	74.44	0.15

^aBias is defined as 100% × (median from bootstrap analysis − estimate from the final model)/estimate from the final model. θ₁, exponent for WT as a covariate for V₁; θ₂, exponent for WT as a covariate for V₂; θ₃, exponent for WT as a covariate for Q; θ₄, exponent for WT as a covariate for V_max; ${\text{ω}}_{V_1}$, square root of the interindividual variance for V₁; ${\text{ω}}_{V_2}$, square root of the interindividual variance for V₂; ω_Q, square root of the interindividual variance for Q; ${\text{ω}}_{V_{\text{max}}}$, square root of the interindividual variance for V_max.

FIG 1.

Comparison of weight-normalized V_max values among three CYP2C19 genotype groups. The solid lines depict the means with standard deviations (SD).

The individual peak concentration (C_max) and C_min at the loading and maintenance doses were estimated according to the individual PK parameters from the final model. The trapezoidal rule was used to calculate the total exposure (area under the concentration-time curve over 24 h [AUC₂₄]) for individuals. To minimize the influence of various doses, 59 cases and 60 cases, respectively, were selected for further group comparisons, depending on the mean values of 6 mg · kg⁻¹ for the loading dose and 4 mg · kg⁻¹ for the maintenance dose. Three groups were divided according to the CYP2C19 phenotype. One-way analysis of variance (ANOVA) and multiple comparisons were employed to assess the differences in C_max, C_min, and AUC₂₄ among three groups. Meanwhile, all of them were normalized by dose and recorded as C_max/D, C_min/D, and AUC₂₄/D, respectively. As shown in Table 4, dose-normalized concentrations (C_max/D and C_min/D) and total exposures (AUC₂₄/D) were significantly different among the three groups, especially between the EM and PM groups.

TABLE 4.

Comparison of dose-normalized concentrations and exposures in patients among the three groups^a

Parameter
	Loading dose (mean, 6 mg/kg)			Maintenance dose (mean, 4 mg/kg)
	Genotype (no. of patients)	Mean ± SD	P	Genotype (no. of patients)	Mean ± SD	P
Cmax/D	EM (22)	0.60 ± 0.06	0.044	EM (24)	0.63 ± 0.08	0.003
	IM (27)	0.62 ± 0.06		IM (25)	0.73 ± 0.22
	PM (10)	0.66 ± 0.06*		PM (11)	0.83 ± 0.12*
Cmin/D	EM (22)	0.07 ± 0.04	<0.001	EM (24)	0.08 ± 0.07	0.012
	IM (27)	0.08 ± 0.04		IM (25)	0.16 ± 0.20
	PM (10)	0.16 ± 0.06*,#		PM (11)	0.24 ± 0.13*
AUC24/D	EM (22)	7.61 ± 0.61	<0.001	EM (24)	8.50 ± 1.64	0.006
	IM (27)	7.81 ± 0.84		IM (25)	10.59 ± 4.99
	PM (10)	8.92 ± 0.74*,#		PM (11)	12.81 ± 2.87*

^a*, P < 0.05, significant difference between poor metabolizer (PM) and extensive metabolizer (EM) groups; #, P < 0.05, significant difference between PM and intermediate metabolizer (IM) groups.

Model evaluation.

As shown in Fig. 2, in goodness-of-fit plots, both the individual prediction (IPRED) and population prediction (PRED) were close to the observed concentration, which responded to an acceptable prediction accuracy of the final model. Most conditional weighted residuals (CWRES) laid within ±2 ranges. The good stability of the final model was confirmed by the results of the bootstrap analysis shown in Table 3. All parameters obtained from the final model were similar to the median values of bootstrap estimates with a bias of ±13%. The visual predictive check (VPC) is shown in Fig. 3. Almost all observed concentrations were within the 95% confidence intervals (CIs) of prediction, indicating the reliable predictive performance of the final model. Furthermore, a good fit between the final model and individual data was performed (Fig. 4) by normalized prediction distribution errors (NPDEs), with a mean of 0.067 (standard error [SE] = 0.077) and a variance of 1.148 (SE = 0.12). The P values were 0.378, 0.153, 0.940, and 0.459, obtained by a t test, Fisher’s variance test, a Shapiro-Wilks test, and a globally adjusted test, respectively. The statistical results suggested that NPDEs followed a normal distribution with variance homogeneity.

FIG 2.

Goodness-of-fit plots of the final model. (a) Observed concentration (DV) versus individual prediction (IPRED); (b) DV versus population prediction (PRED); (c) conditional weighted residuals (CWRES) versus PRED; (d) CWRES versus time. In panels a and b, the black line is the uniform line, and the red line is the regression line. The tendency curves of conditional weighted residuals are shown in panels c and d.

FIG 3.

Visual predictive check of the final model. The solid lines represent the median simulated concentrations (black lines) and median observed concentrations (red lines). The 2.5th and 97.5th percentiles of them are presented as dashed lines. The three shaded areas represent the 95% intervals of the three black lines.

FIG 4.

NPDE verification plots of the final model. (a and b) Quantile-quantile plot (a) and histogram (b) of the normalized prediction distribution errors (NPDEs). (c) NPDE versus time. (d) NPDE versus population prediction (PRED).

Optimization of the dosing regimen.

As mentioned above, V_max is influenced by various variables, including WT, CYP2C19 phenotype, and omeprazole. For patients in every subpopulation, different dosing regimens were simulated to achieve the target concentration of 1.0 to 5.5 μg · ml⁻¹. The estimated trough concentrations and probability of target attainment (PTA) are provided in Tables S1 to S3 in the supplemental material. The simulated concentration-time profiles for each subpopulation are presented in Fig. S1. Meanwhile, the recommended optimal dosing regimens are shown in Table 5. It is suggested that the majority of children should initially be given two loading doses of 9 mg · kg⁻¹ q12h to reach the target concentration in a timely manner. However, a loading dose of 6 to 7.5 mg · kg⁻¹ q8h on day 1 was necessary to improve the initial C_min of young children weighing ≤18 kg, except for those in the PM group. The maintenance doses decreased about 30 to 40% in PM patients compared to EM patients. The maintenance doses were inversely related to WT. The maintenance doses in patients with lower weights could be as high as 9 mg · kg⁻¹ q12h. Furthermore, the maintenance doses decreased slightly when coadministered with omeprazole.

TABLE 5.

Optimal dosage regimens for patients with different weights, CYP2C19 phenotypes, and coadministrations with omeprazole

WT (kg)	CYP2C19 phenotypea	Without omeprazole		With omeprazole
		Loading dose (mg · kg−1), frequency	Maintenance dose (mg · kg−1)	Loading dose (mg · kg−1), frequency	Maintenance dose (mg · kg−1)
9	EM	7.5, q8h	9	6.5, q8h	7
	IM	6.5, q8h	8	9, q12h, or 6, q8h	6
	PM	9, q12h, or 6, q8h	6	9, q12h, or 6, q8h	5
18	EM	6.5, q8h	8	9, q12h, or 6, q8h	6
	IM	6, q8h	6.5	9, q12h, or 6, q8h	5.5
	PM	9, q12h, or 6, q8h	5	9, q12h, or 6, q8h	4
36	EM	9, q12h, or 6, q8h	6.5	9, q12h, or 6, q8h	5
	IM	9, q12h, or 6, q8h	5.5	9, q12h, or 6, q8h	4.5
	PM	9, q12h, or 6, q8h	4	9, q12h, or 6, q8h	3

^aEM, extensive metabolizer; IM, intermediate metabolizer; PM, poor metabolizer.

DISCUSSION

To the best of our knowledge, population PK studies of VRC in Chinese pediatric patients are limited. In the current study, a population PK analysis of VRC in Chinese children with critical illness was successfully carried out. Different from adult populations, the PK data in our study were well described by a two-compartment model with nonlinear elimination, which was often employed in pediatric patients in previous reports (4, 6, 10). Due to a lack of serum concentrations within 24 h after the initial dose, a time-dependent model was not conducted in the current study. WT, CYP2C19 phenotype, and omeprazole were found to significantly affect the V_max of VRC. After normalization with WT, the estimated V_max was 1.007 mg · h⁻¹ · kg⁻¹, which is approximately equal to the previously reported median value (0.957 mg · h⁻¹ · kg⁻¹) (4) but higher than the V_max (0.736 mg · h⁻¹ · kg⁻¹) in German children (10). This difference may be attributed to the distribution of age ranges.

Age-associated changes are dynamic in body composition and organ function (11). In adult populations, age is included in PK models as a significant variable (2). However, for young children, WT is highly correlated with age, reflects the extent of allometric maturation, and seems to be more important than age in predicting PK characteristics (6, 12). Our study also confirmed that WT is superior to age as a primary covariate by using allometric models (Table 2).

A retrospective analysis (13) indicated that increased C-reactive protein (CRP) and decreased albumin (ALB) could alter the concentration/dose ratio, suggesting a reduction in VRC elimination, but CRP and ALB were not included in our final model due to the slight changes in the OFVs (ΔOFVs, 1.56 and 0.58). For the same reason, aspartate aminotransferase (AST), interleukin-6 (IL-6), and glucocorticoids were also excluded, which have also been reported to be related to VRC trough concentrations (13, 14).

VRC is mainly metabolized in the liver by the drug-metabolizing enzyme CYP2C19, while CYP3A4 and CYP2C9 contribute less to its metabolism (15). The extent of CYP2C19 activity, reflected by genotyping (UM, EM, IM, and PM), is highly associated with the elimination velocity of VRC and included in our final model as an important covariate. Statistically, the incidence of PM is higher in Chinese individuals (14.7%) than in Caucasians and Africa (2.1% and 3.7%) (16), while the mutant CYP2C19*17 allele has a lower frequency in Asians (<2%) than in Caucasians (21.3%) (2). In the present study, the incidences of PM and UM were 14.1% and 1%, respectively, which approximately corresponded to data in previous reports. There was one patient who provided a mutant CYP2C19*2/*17 allele, which is too rare to confirm its activity. Other studies have revealed that this is still not certain and needs further analysis to determine whether the influence of the CYP2C19*2/*17 genetic variant on VRC PK properties is close to the CYP2C19*1/*17 genotype or the IM group (17, 18). In our study, the limited number of subjects with a *17 variant excluded them from group comparisons. In addition, another gene, ABCB1, also known as MDR1 or P-glycoprotein (P-gp), which could contribute to altering the clearance of VRC and to predicting VRC serum concentrations (19, 20), was not included in the final model since there was no significant change in the OFV (ΔOFV, −0.37).

The CYP2C19 genotype, as a covariate on V_max, helps to explain the interindividual variabilities in VRC concentrations and exposures. Other clinical investigators have revealed that the V_max or clearance of VRC in PM patients was significantly lower than that in IM and EM patients, while the serum exposure level was greatly increased (4, 12, 21). Figure 1 shows a roughly downward trend in the weight-normalized V_max from EM to PM patients, which confirmed the results from previous studies. Lee et al. found that IM and PM patients had 1.5- and 3.4-fold-higher AUCs than their EM counterparts and had 2.8- and 5.1-fold-higher C_min values than their EM counterparts, respectively (22). In the present study, significant differences in concentrations and exposures were also observed between the EM and PM groups (Table 4). Regardless of coadministered medicine, the dose-normalized C_max, C_min, and AUC₂₄ in PM patients were approximately 1.1-, 2.3-, and 1.2-fold higher than those in EM patients under a loading dose of 6 mg · kg⁻¹, while they were instead 1.3-, 3.0-, and 1.5-fold higher under a maintenance dose of 4 mg · kg⁻¹. However, there were no significant changes between the EM and IM groups.

Despite the weak evidence of the relationship between genotype and clinical efficacy, adjustment of the initial dosage guided by genotype still has potential advantages (21). Various guidelines suggest that the VRC dosing schedule should be adjusted according to the patient’s CYP2C19 phenotype, but these guidelines do not recommend more detailed information about the dosing schedule (15, 16, 23). In the present study, different dosing regimens were simulated with regard to different CYP2C19 phenotypes (see Tables S1 to S3 and Fig. S1 in the supplemental material). Our study found that the current dose of 4 mg · kg⁻¹ was insufficient to reach the target concentration, especially for children with the EM genotype. Walsh et al. revealed that exposure of 4 mg · kg⁻¹ in children was equal to that of only 3 mg · kg⁻¹ in adults (12). Similar to previous reports (7, 24), 57.5% of measured trough concentrations were at subtherapeutic levels. According to Bayesian estimates, a total of 67.7% of patients were expected to require increased doses. For children with the EM phenotype, a maintenance dose of 5 to 9 mg · kg⁻¹ q12h is required. However, children weighing >36 kg with the PM phenotype may have an increased risk of a C_min of >5.5 μg · ml⁻¹ after 5 days of VRC therapy following the previously reported maintenance dose of 8 mg · kg⁻¹ q12h (Fig. S1c). Moreover, the dosage is increased in young patients with lower weights, which might be associated with their higher weight-normalized V_max and is consistent with the suggestion of a previous study (7). As it was reported that a loading dose of 9 mg · kg⁻¹ i.v. in children was equivalent to 6 mg · kg⁻¹ i.v. in adults (6), the simulated results suggested that two loading doses of 9 mg · kg⁻¹ were reliable for most children to achieve the 50% PTA of a C_min of ≥1 μg · ml⁻¹. But for young children weighing ≤18 kg in the EM and IM groups, the loading dose on day 1 should be improved to 6 to 7.5 mg · kg⁻¹ q8h to achieve comparable exposures.

At present, there are limited data on VRC therapy in children aged <2 years. Recommending an appropriate dosing regimen for this special population remains a challenge. Liu et al. suggested that Chinese children aged <2 years could have an increased chance of reaching the therapeutic target following a maintenance dose of 5 to 7 mg · kg⁻¹ twice daily (25). Bartelink et al. observed a higher daily dose of 12 to 71 mg · kg⁻¹ for 11 children in this age range (7). In the current study, there were eight children aged <2 years, with a median weight of 10 kg. The recommended optimal dosing regimens for this cohort are shown in Table 5. When administered without omeprazole, the dosages of 6 mg · kg⁻¹ and 8 mg · kg⁻¹ q12h were adequate for patients in the PM and IM groups, respectively, whereas the EM patients required a dosage of 9 mg · kg⁻¹ q12h. However, due to the limited sample size, the recommendations are only for reference, and further clinical practice is required.

The limitations of our study should be mentioned. The concentration data from oral dosing were not investigated, resulting in the lack of exploration of the absorption rate and oral bioavailability. A pharmacokinetic-pharmacodynamic model was not constructed because of the paucity of data on galactomannan (many patients have only one observation). In addition, the CYP2C19*17 allele is still expected to be explored by richer data sets.

Conclusion.

In the present study, we conducted a population PK model of VRC in Chinese children with critical illness. The results suggested that weight, CYP2C19 genotype, and omeprazole have a significant influence on the V_max of voriconazole. The CYP2C19 genotype contributed to explaining the high PK variability, and it is necessary to adjust dosing regimens according to the CYP2C19 genotype. Optimal dosing regimens have been recommended for Chinese children based on the final model.

MATERIALS AND METHODS

Patients.

All children with critical illness receiving VRC for more than 3 days were eligible for inclusion within the Hematology Department and Intensive Care Unit at Wuhan Children’s Hospital (Hubei, China) from May 2019 to October 2020. Patients with incomplete information were excluded. VRC was used for treatment or antifungal prophylaxis in children with IFDs, who were identified according to Chinese guidelines (26) and European Organization for Research and Treatment of Cancer and Mycoses Study Group (EORTC/MSG) criteria (27). VRC was administered as an intravenous infusion over 1 h. Dose adjustment was decided by physicians according to clinical conditions. An opportunistic sampling strategy was performed at least 30 h after the start of therapy. The remaining blood after routine biochemical tests was collected and determined. Demographic, laboratory, and clinical data were collected from the electronic medical record database in the hospital. The VRC dosing regimens and serum concentrations for at least one sample per patient were recorded. Concomitant drugs that could affect the PKs of VRC, including omeprazole and glucocorticoids, were also recorded.

The Ethics Committee of Wuhan Children’s Hospital approved this study according to legal requirements, and informed consent was obtained from the parents or guardians of individual patients.

Assay of serum voriconazole.

VRC in serum was extracted by a Cleanert ODS C₁₈-solid-phase extraction (SPE) column (200 mg/3 ml) and detected by high-performance liquid chromatography (1260 infinity II; Agilent Technologies Inc., Santa Clara, CA) with UV detection. The analytical column was a SinoChrom ODS-BP column (4.6 by 250 mm, 10 μm; Agela Technologies), and the mobile phase was methanol-acetonitrile-ammonium acetate buffer (25 mmol · liter⁻¹, pH 4.6) at 20:32:48 (vol/vol/vol) with a flow rate of 0.8 ml · min⁻¹. The detection wavelength was set to 256 nm, and the temperature was 30°C. A good liner calibration curve of VRC was obtained in the range of 0.26 to 16.70 μg · ml⁻¹ (r = 0.999), and the limit of detection was 0.03 μg · ml⁻¹ (signal-to-noise [S/N] ratio of ≥3). Intra- and interday relative standard deviation (RSD) values were less than 10%.

Genotyping.

DNA was extracted from venous blood leukocytes with the traditional phenol-chloroform method. Genotyping for CYP2C19*2, CYP2C19*3, and CYP2C19*17 alleles and ABCB1 was performed using Sanger sequencing by Sangon Biotech (Shanghai, China). CYP2C19 genotyping was as follows: ultrarapid metabolizer (UM) (CYP2C19*17/*17 and -*1/*17), extensive metabolizer (EM) (CYP2C19*1/*1), intermediate metabolizer (IM) (CYP2C19*1/*2 and -*1/*3), and poor metabolizer (PM) (CYP2C19*2/*2, -*2/*3, and -*3/*3) (2).

Population PK model development.

Phoenix NLME (version 8.2.0; Pharsight Corporation, CA, USA) was used to analyze the population PK data of VRC. Output visualization and model validation were performed with the R program (version 3.5.1; https://www.r-project.org/). The first-order conditional estimation, extended-least-squares method was employed during model development.

Base model.

One- and two-compartment models with linear or saturable elimination were compared to fit the VRC concentration-time data. The structural model was evaluated according to visual inspection of routine diagnostic plots as well as the objective function value (OFV), Akaike information criterion (AIC) value, and Bayesian information criterion (BIC) value.

For the PK parameters, interindividual variability was expressed as an exponential model with a mean of zero and a variance of ω², while residual-error variability was described by a proportional model with a mean of zero and a variance of σ².

Covariate analyses.

Covariates were selected or excluded by forward-inclusion and backward-elimination stepwise procedures with a likelihood ratio test. A covariate was considered significant for inclusion if a reduction in the OFV of >3.84 (P < 0.05) was obtained in the forward-inclusion step and an increase in the OFV of >6.63 (P < 0.01) was achieved in the sequential backward-elimination step. The final model with the lowest OFV, AIC value, and BIC value was regarded as superior. An empirical Bayesian method was employed to estimate the individual PK parameters based on the final model.

Various variables, including age, sex, height, weight (WT), body surface area (BSA), alanine aminotransferase level, aspartate aminotransferase (AST) level, total bilirubin level, albumin (ALB) level, blood urea nitrogen level, serum creatinine concentration, uric acid level, estimated glomerular filtration rate (eGFR), C-reactive protein (CRP) level, interleukin-6 (IL-6) level, genotype, and concomitant drugs, were investigated to assess their potential influence on the PK parameters. The BSA and eGFR were calculated according to the Mosteller formula (28) and the updated Schwartz formula (29), respectively. In this process, continuous covariates and categorical covariates were included in the forms of the power model and the exponential model, respectively.

It is well known that developmental factors play an important role in the metabolism of VRC in children. Therefore, six candidate models (summarized in Table 2) were tested to describe differences in allometric size and the process of clearance.

To describe the reduction in the maximum velocity of elimination (V_max) over time, a time-dependent model (model VII) was tested subsequently, according to a previous study (6), as follows:

$$V_{\text{max}}=V_{\text{max},1}\times \left(1-V_{\text{max},\text{inh}}\times \frac{\left(t-1\right)}{\left(t-1\right)+\left(t_{50}-1\right)}\right)$$ (5)

where V_max,1 represents the maximum elimination rate at 1 h and V_max,inh represents the maximum fraction of inhibition. t₅₀ describes the time in hours after the initiation of dosing where half-maximum inhibition occurred.

Model validation.

For initial diagnosis, goodness-of-fit plots were performed in terms of the observed concentration (DV) versus individual prediction (IPRED), DV versus population prediction (PRED), conditional weighted residuals (CWRES) versus PRED, and CWRES versus time. One thousand resamplings with replacement were produced by a bootstrap approach, and estimated median parameters with 95% confidence intervals (CIs) were calculated and compared with the parameters obtained from the final model. To evaluate the prediction performance of the final model, a visual predictive check (VPC) was performed by comparing the observations and simulations. The final model was further assessed graphically and statistically by normalized prediction distribution errors (NPDEs).

Optimization of the dosing regimen.

Based on the PK parameters from the final model, the C_min values of various dosing regimens were estimated for each subpopulation stratified by WT (9 kg, 18 kg, and 36 kg), CYP2C19 phenotype (EM, IM, and PM), and coadministration with omeprazole. The probability of target attainment (PTA) of a C_min of between 1.0 and 5.5 μg · ml⁻¹ was calculated to recommend an optimal dosing regimen for each subpopulation (23, 25). A VRC concentration-time curve was also simulated for each subpopulation, following two loading doses of 9 mg · kg⁻¹ q12h and a maintenance dose of 8 mg · kg⁻¹ q12h.