Population Pharmacokinetic Study of Prophylactic Fluconazole in Preterm Infants for Prevention of Invasive Candidiasis

Fluconazole is an antifungal agent with reported evidence for its prophylactic effect against systemic fungal infection in preterm infants. The aim of this study was to build a population pharmacokinetic model to evaluate the pharmacokinetic characteristics of intravenous and oral fluconazole in preterm infants with the current prophylactic fluconazole dosing regimen.

ABSTRACT

Fluconazole is an antifungal agent with reported evidence for its prophylactic effect against systemic fungal infection in preterm infants. The aim of this study was to build a population pharmacokinetic model to evaluate the pharmacokinetic characteristics of intravenous and oral fluconazole in preterm infants with the current prophylactic fluconazole dosing regimen. A pharmacokinetic model was developed using 301 fluconazole concentrations from 75 preterm infants with a baseline body weight (WT) ranging from 0.5 to 1.5 kg and an estimated glomerular filtration rate (eGFR) ranging from 12.9 to 58.5 ml/min/1.73 m². Eligible infants received an intravenous or oral dose of 3 mg/kg of body weight of fluconazole, twice weekly with a ≥72-h dose interval, for 4 weeks. The model was qualified with basic goodness-of-fit diagnostics, visual predictive checks, and bootstrapping. The fluconazole pharmacokinetics was well described with a one-compartment linear model with a proportional residual error. The population clearance (CL) and volume of distribution (V) were derived as 0.0197 × (WT/1.00)^0.746 × (eGFR/25.0)^0.463 × exp(η) and 1.04 × WT × exp(η), respectively. Such covariate analyses augment the awareness of the need for personalized dosing in preterm infants. (This study has been registered at ClinicalTrials.gov under identifier NCT01683760.)

INTRODUCTION

Prophylactic uses of antifungal agents for the prevention of systemic fungal infections are used in both adult patients with immunodeficiency or cancer and preterm infants with immature immune systems (1–5). Such uses of antifungal agents do not necessarily eliminate infections but reportedly decrease the number of events of invasive systemic candidiasis, leading to a reduction in overall mortality by the suppression of Candida colonization (6–12). The widely used classes of antifungal agents for the prevention of systemic candidiasis are triazoles and polyenes (7, 13).

Numerous prophylactic uses of fluconazole, a synthetic triazole, have been reported to reduce the rate of invasive candidiasis in preterm infants with extremely low birth weight (ELBW) via the suppression of Candida colonization (6, 9, 11, 14, 15). Out of the pharmacokinetic-pharmacodynamic (PK-PD) indices for measuring antifungal activity, the ratio of the fluconazole dose to the MIC, termed the dose/MIC ratio, and the ratio of the area under the curve over 24 h after dose at steady state (AUC₂₄) to the MIC, termed the AUC/MIC ratio, are known to best reflect the antifungal activity of fluconazole (16–20). In adults, maintained exposure with a daily dose of 100 mg was considered equivalent to an AUC₂₄ of 100 mg · h/liter; the AUC/MIC and dose/MIC ratios give rise to identical numerical values (19). A minimum efficacious exposure of fluconazole can be estimated based on the increased success rate of therapeutic response and a dose/MIC ratio of >50 among adult patients with candidemia (19, 21). For fluconazole-susceptible Candida species based on the Clinical and Laboratory Standards Institute (CLSI) sensitivity breakpoint at an MIC of ≤8 mg/liter, the predicted minimum efficacious exposure of fluconazole required to ensure at least 70% efficacy for the treatment of invasive systemic candidiasis to meet an AUC/MIC ratio of >50 corresponds to an AUC₂₄ of 400 mg · h/liter (21–25). This is equivalent to the susceptible dose-dependent species with MIC breakpoints of 32 mg/liter to meet an AUC/MIC ratio of >12, which is often cited as efficacious exposure for adults (26).

A daily dose of 400 mg is recommended in drug labeling for adults undergoing bone marrow transplantation requiring fluconazole prophylaxis; however, retrospective analyses from randomized controlled trials (RCTs) have shown that reduced doses of 100 to 200 mg gave a similar incidence of breakthrough infections (27, 28). When the dose equivalency scheme from adults is applied, daily doses of 100 and 200 mg resulting in an AUC₂₄ of 100 to 200 mg · h/liter would correspond to the susceptible dose-dependent MICs of 2 and 4 mg/liter, respectively, for an AUC/MIC ratio of >50 (23). Due to the limited experience of PK-PD studies in preterm infants receiving fluconazole for preventive purposes, such presumptions are inevitable, and observational evidence remains to be provided (9). Thus far, the suggested dosing regimens for the early prevention of invasive systemic candidiasis in infants from Monte Carlo simulations of 3 and 6 mg/kg of body weight twice weekly are expected to meet AUC/MIC ratios of >50 for MICs of 2 and 4 mg/liter by maintaining the fluconazole concentration above 2 mg/liter for more than 40% during a ≥72-h interval with an AUC₂₄ of 50 to 100 mg · h/liter (8, 19).

Although numerous findings support the effectiveness of fluconazole prophylaxis through RCTs, there are also controversial affirmations with debate on whether it truly reduces the mortality and morbidity rates among very-low-birth-weight (VLBW) infants with a birth weight (WT) of ≤1.5 kg (11, 12, 14, 15, 29–31). The link between fluconazole prophylaxis and its efficacy is still missing, and whether the lack of efficacy in systemic candidiasis prevention is due to other factors, such as insufficient exposure required to elicit the prophylactic effect, still remains unanswered. Therefore, a personalized approach through studies on the PK characteristics of fluconazole prophylaxis in the preterm infant population is necessary, because such information cannot be extrapolated from adults or full-term pediatric data (25).

Based on these understandings, this study aims to characterize the fluconazole PK in VLBW infants who have received fluconazole prophylaxis therapy, through population PK (PopPK) model building and evaluation.

RESULTS

Data set and baseline subject demographics.

Eighty eligible preterm infants were enrolled, with five in the intensive sampling group; however, five infants were excluded from the present study due to withdrawal of consent prior to study drug administration. In total, 303 fluconazole plasma concentrations from the remaining 75 preterm infants (with 3 in the intensive sampling group) were included in the population PK analysis (see Fig. SA1 in the supplemental material). The baseline median (interquartile range) values for gestational age (GA), postmenstrual age (PMA), postnatal age (PNA), WT, and estimated glomerular filtration rate (eGFR) were 30.3 (28.0 to 31.3) weeks, 30.6 (28.4 to 31.6) weeks, 3.0 (3.0 to 4.0) days, 1.1 (0.9 to 1.3) kg, and 26.7 (22.9 to 32.3) ml/min/1.73 m², respectively (Table 1).

TABLE 1.

Baseline demographic characteristics of the subjects

^aData are medians (interquartile ranges) or frequencies (percentages) unless otherwise specified.

^bEstimated using the bedside Schwartz formula for infants, eGFR = 0.413 × HT/CRE.

Population pharmacokinetic analysis.

A one-compartment linear model structure with the bioavailability (F) estimate for the oral and intravenous administration of fluconazole adequately described the concentration-time profiles in preterm infants (Table 2 and Fig. SA2). The proportional error model included the between-subject variability (BSV) for the clearance (CL), the volume of distribution (V), and the first-order absorption rate constant for oral dose (K_a). The effect of WT on the CL and V was selected with estimated exponents and simple allometry, respectively. As a result of numerous covariate screenings, creatinine (CRE), GA, PMA, and PNA were ruled out from the significant covariates for CL and V (Table 2). Meanwhile, the exponential model described the gradual increase of CL; eGFR in the power term was a significant covariate for CL (Fig. 1 and Table 3) (32). In general, WT and eGFR had a positive relationship with the fluconazole CL, as expected.

TABLE 2.

Model-building process^a

Model	New model	Model description	Population model	OFV	ΔOFV
Reference model
—	100	1-compartment	CL = θCL × eη1	306.759
		WT on V	V = θV × (WT/1.00) × eη2
100	101	2-compartment	CL = θCL × eη1	306.758	−0.001
			V2 = θV2 × (WT/1.00) × eη2
		WT on V	V3 = θV3 × (WT/1.00) × eη3
100	102	1-compartment, IIV on CL	CL = θCL × eη1	152.974	−153.785
		WT on V	V = θV × (WT/1.00) × eη2
100	103	1-compartment, IIV on V	CL = θCL × eη1	225.421	−51.338
		WT on V	V = θV × (WT/1.00) × eη2
100	104	1-compartment, IIV on Ka	CL = θCL × eη1	288.064	−18.695
		WT on V	V = θV × (WT/1.00) × eη2
102	105	1-compartment, IIV on CL, V	CL = θCL × eη1	126.160	−26.814
		WT on V	V = θV × (WT/1.00) × eη2
102	106	1-compartment, IIV on CL, Ka	CL = θCL × eη1	114.974	−38.000
		WT on V	V = θV × (WT/1.00) × eη2
103	107	1-compartment, IIV on V, Ka	CL = θCL × eη1	242.456	−12.965
		WT on V	V = θV × (WT/1.00) × eη2
106	108	1-compartment, IIV on CL, V, Ka	CL = θCL × eη1	96.156	−18.818
		WT on V	V = θV × (WT/1.00) × eη2
108	109	1-compartment, IIV on CL, V, F1	CL = θCL × eη1	126.033	+29.877
		WT on V	V = θV × (WT/1.00) × eη2
108	110	1-compartment, IIV on CL, V, Ka, F1	CL = θCL × eη1	96.156	0
		WT on V	V = θV × (WT/1.00) × eη2
Base model		1-compartment, IIV on CL, V, Ka	CL = θCL × eη1
		WT on V	V = θV × (WT/1.00) × eη2
Reference model
108	8001	WT on V	V = θV × (WT/1.00) × eη2	96.156
8001	8002	WT on V	CL = θCL × (eGFR/25)θeGFR × eη1	46.094	−50.062
		eGFR on CL	V = θV × (WT/1.00) × eη2
8001	8003	WT on V, CL	CL = θCL × (WT/1.00)θWT × eη1	44.430	−51.726
			V = θV × (WT/1.00) × eη2
8003	8004	WT on V	CL = θCL × (WT/1.00)θWT × (GA/30)θGA × eη1	42.148	−2.282
		WT, GA on CL	V = θV × (WT/1.00) × eη2
8003	8005	WT on V	CL = θCL × (WT/1.00)θWT × (PNA/5)θPNA × eη1	34.770	−9.660
		WT, PNA on CL	V = θV × (WT/1.00) × eη2
8003	8006	WT on V	CL = θCL × (WT/1.00)θWT × (PMA/30)θPMA × eη1	30.282	−14.148
		WT, PMA on CL	V = θV × (WT/1.00) × eη2
8003	8007	WT on V	CL = θCL × (WT/1.00)θWT × (eGFR/25)θeGFR × eη1	17.211	−27.219
		WT, eGFR on CL	V = θV × (WT/1.00) × eη2
8007	8008	WT on V	CL = θCL × (WT/1.00)θWT × (eGFR/25)θeGFR × (GA/30)θGA × eη1	14.251	−2.960
		WT, eGFR, GA on CL	V = θV × (WT/1.00) × eη2
8007	8009	WT on V	CL = θCL × (WT/1.00)θWT × (eGFR/25)θeGFR × (PNA/5)θPNA × eη1	17.491	+0.280
		WT, eGFR, PNA on CL	V = θV × (WT/1.00) × eη2
8007	8010	WT on V	CL = θCL × (WT/1.00)θWT × (eGFR/25)θeGFR × (PMA/30)θPNA × eη1	13.930	−3.281
		WT, eGFR, PMA on CL	V = θV × (WT/1.00) × eη2
8003	8011	WT on V	CL = θCL × (WT/1.00)θWT × (CRE/0.6)θeGFR × eη1	44.430	0
		WT, CRE on CL	V = θV × (WT/1.00) × eη2
Final model		WT on V	CL = θCL × (WT/1.00)θWT × (eGFR/25)θeGFR × eη1
		WT, eGFR on CL	V = θV × (WT/1.00) × eη2

^aAbbreviations: CL, clearance; IIV, interindividual variability; WT, weight; eGFR, estimated glomerular filtration rate; PNA, postnatal age in days; GA, gestational age in weeks; PMA, postmenstrual age in weeks; V, volume of distribution; OFV, objective function value; ΔOFV, change in the OFV compared to the reference model; e^η1, exp(η₁).

FIG 1.

Schematic diagram of the final model. PO, oral; IV, intravenous.

TABLE 3.

Parameter estimates of the final population pharmacokinetic model of prophylactic fluconazole in preterm infants^a

Parameter	Description (unit)	Estimation result {estimate [%RSEb ] or (%CV)}	Bootstrap resultc [median value (95% CI)]
Ka = θ1
θ1	First-order absorption rate constant (1/h)	0.538 [18.5]	0.536 (0.389, 0.729)
CL = θ2 × (WT)θ7 × (eGFR/25)θ8 × exp(η2)
θ2	Clearance (liters/h)	0.0197 [6.99]	0.0194 (0.0174, 0.0217)
θ7	Change in CL due to WT difference (liters/h)	0.746 [29.2]	0.741 (0.318, 1.004)
θ8	Change in CL due to eGFR difference (liters/h)	0.463 [24.4]	0.456 (0.281, 0.683)
V = θ3 × (WT) × exp(η3)
θ3	Vol of distribution (liters)	1.04 [8.23]	1.02 (0.882, 1.15)
F = θ4
θ4	Relative bioavailability of first-order absorption (%)	0.909 [7.03]	0.896 (0.806, 0.986)
Interindividual variance
ω12	BSV of Ka	1.44 (28.2)	1.40 (0.758, 2.11)
ω22	BSV of CL	0.0552 (33.0)	0.0502 (0.0239, 0.0840)
ω32	BSV of V	0.0447 (51.6)	0.0403 (0.0111, 0.0808)
Residual variance
σ22	Residual error (proportional)	0.282 (7.61)	0.279 (0.242, 0.312)

^aAbbreviations: BSV, between-subject variability; %CV, coefficient of variation expressed as a percentage; CI, confidence interval; CRE, creatinine level; eGFR, estimated glomerular filtration rate; RSE, relative standard error.

^b%RSE = (standard error/parameter estimate) × 100.

^cDetermined from a parametric bootstrap resampling of 500 replicates.

The goodness-of-fit (GOF) and visual predictive check (VPC) plots and the bootstrap resampling of the final model showed good adequacy between the observed and predicted fluconazole concentrations obtained from preterm infants (Fig. 2 and 3). The final model parameter estimates had good precision, as presented in Table 3, with the 95% confidence intervals (CIs) generated by the bootstrap-resampled simulated trials. The percent relative standard error around the parameter point estimates was in the range of 7.0 to 29.2%, along with the maximum interindividual variance of 51.6% for V, shown as the percent coefficient of variation (%CV). Overall, the final PK model was robust at describing the PK characteristics of oral and intravenous fluconazole administrations in VLBW preterm infants.

FIG 2.

Basic goodness-of-fit plot of the final pharmacokinetic model. Solid lines, line of identity; dotted lines, line of reference (y = 0); gray lines, line of locally weighted scatterplot smoothing (LOWESS).

FIG 3.

Visual predictive check results for oral (a) and intravenous (b) administrations. Solid red lines, median from the time-concentration profile at steady state; solid blue lines, lower and upper limits of the 95% confidence interval; shaded areas in red, 95% confidence intervals for the median of the simulated concentrations; shaded areas in blue, 5th and 95th percentiles of the simulated concentrations. All y axes are presented in linear scales.

DISCUSSION

The controversy over the effectiveness of fluconazole prophylaxis in preterm infants remains to be followed up with a further understanding of the relationship between the PK, proposed PD indices, and clinical outcomes. In underobserved populations such as preterm infants, a population pharmacokinetic approach is a known stepping-stone toward personalized pharmacotherapy. In this study, we have constructed an adequate PK model for preterm infants who receive a routine prophylactic fluconazole regimen (Table 3 and Fig. 1).

The goal for dose selection in infant populations primarily lies upon understanding the sources of within- and between-subject variabilities in terms of their PK characteristics. The major candidate sources of PK variability are within growth and maturation in infants, contributing order-of-magnitude differences in body weight compared to adults (33, 34). In a study by Rhodin et al., WT and PMA were used as the maturation factors to describe human renal function (35). Therefore, to disentangle competing covariate influences, WT was used as a surrogate for growth as the first covariate to explain the interindividual differences in a systemic approach (36, 37). The magnitude of power for the CL model was estimated rather than using fixed time-related or physiological variables scaled with a WT exponent of 0.25 or 0.75, respectively (38). The estimated power term within the population for CL (0.746) was almost identical to the allometric 3/4 power model when the eGFR was accounted for as the covariate. Unlike the population PK model previously reported by Wade et al., where GA and PNA were used to describe CL in infants with a WT range of 0.5 to 7.1 kg, the birth WT was included as the model’s most significant covariate (Table 2) (39). No other surrogate for maturation, GA, PNA, or PMA, further influenced the CL of this population, which consequently implies that in VLBW infants, like this population (premature infants with birth WT of ≤1.5 kg), the increase in WT influences the maturation of the kidney, which has the largest effect on the CL of fluconazole (40). The second covariate used to explain the fluconazole CL in the study infants was the eGFR calculated with the original bedside Schwartz formula for children (41). Considering the variables height (HT) and CRE included in the formula, HT may seem redundant for the effect of growth along with WT, which has already been included as a covariate. However, when CRE was examined within the covariate analyses, a similar improvement in the model fit was observed (Table 2). Therefore, it seems feasible to assume that there is no correlation between WT and CRE; hence, they are two independent covariates.

Fluconazole is used as a marker for passive distal tubular reabsorption due to its extensive tubular reabsorption (42, 43). In terms of glomerular filtration, the acquisition of renal function is completed during postnatal development, whereas tubular function matures more progressively with greater efficiency in infants for some mechanisms than in adults (44, 45). Drug transporter expression is also known to differ in an age-dependent manner for hepatic and intestinal drug transporters (46). Therefore, the contribution of immature passive renal drug transporter compositions can lead to a lesser extent of reabsorption of fluconazole into the system, which may contribute to the insufficient exposure observed compared to that anticipated within the premature infant population (34). Another maturational alteration compared to adults and older children is the decreased plasma proteins in the neonatal population, which may lead to an increase in the unbound drug concentration (47). For drugs such as phenobarbital, the percent difference in protein binding in neonates is close to 50% (48). Collectively, approaches different from those for adults are bound to be applied in preterm infants who are constantly undergoing maturation of the organs, which affects the PK of drugs administered through personalized optimum pharmacotherapy.

One drawback of the present study is that significant covariates for the CL of fluconazole in the neonate population within this study were WT and eGFR, and the weight range was 0.5 to 1.5 kg, with a maximum calculated eGFR of 58.5 ml/min/1.73 m². In a previous population PK study in young infants with a WT range of 0.5 to 7.1 kg, fluconazole CL was best described by WT, CRE, PNA, and GA at birth (39). Another population PK study in extremely-low-birth-weight (ELBW) infants with a WT of ≤0.75 kg found that CRE was the only predictor of fluconazole CL (49). The differences between the selected covariates and those previous studies may be due to the narrower ranges of candidate covariates for the observed population and the WT range of the subjected infants included within the analyses. Second, not only WT and HT but also CRE data were not an exact match for each dosing and PK sampling time point. In addition, we included only 75 infants in the analysis. However, the PK parameters derived from the final model were in agreement with those derived by Wade et al. (39). Addressing these concerns would require further study with a larger sample size. Third, although the original bedside Schwartz formula is a validated method for the estimation of GFR in a predominantly nonchronic kidney disease population regardless of birth weight or term age due to the direct impact of the maternal CRE on the serum CRE in infants during the first 14 days of life, substantial limitations on the estimation of GFR lie within (41, 50). In addition, cystatin C and beta-trace protein are known to reflect neonatal renal function more accurately, with minimal amounts crossing the placenta, unlike CRE (51). Neither cystatin C nor beta-trace protein levels were available in the present study, hence the use of HT and CRE for the eGFR calculation. However, such a drawback can also be seen as an advantage, in that the eGFR with the Schwartz equation can easily be calculated in clinical practice with a routine safety laboratory item and growth measurements.

With the final PK model adequately describing the observed plasma concentration of prophylactic fluconazole in the preterm infant population, this study suggests that WT at birth and eGFR are predictors for fluconazole CL in the observed population. Such work is expected to be useful for the further evaluation of the PK-PD indices proposed thus far for prophylactic fluconazole use in preterm infants, complemented with observations of long-term adverse effects, such as possible concerns of adrenocortical insufficiency, neurodevelopment, and hyperbilirubinemia (52–54).

MATERIALS AND METHODS

Study design.

This study (ClinicalTrials.gov identifier NCT01683760) was conducted in full accordance with the principles stipulated in the Declaration of Helsinki as amended in 2008 (Seoul, Republic of Korea) (55) and International Conference on Harmonization (ICH) guidelines on good clinical practice (GCP) (56). The study protocol was reviewed and approved by the Seoul National University Hospital Institutional Review Board (IRB) (IRB no. H-1207-024-416).

A prospective, single-center, open-label, pharmacokinetic study was performed in preterm infants. Voluntary informed written consent for participation in this study was obtained from the legal guardians of the preterm infants hospitalized at the neonatal intensive care unit (NICU) at Seoul National University Children’s Hospital who required fluconazole prophylaxis for the prevention of systemic fungal infection and who had birth weights of ≤1.5 kg. Infants who received an antifungal agent for treatment purposes or with biological mothers who had been treated with an antifungal agent due to intrauterine fungal infection during pregnancy were excluded from the study.

Starting from the third day after birth, 80 eligible preterm infants received 3 mg/kg fluconazole (Diflucan; Pfizer Inc., Seoul, Republic of Korea), orally through an orogastric tube or intravenously if a venous catheter was present, with at least a 72-h interval twice weekly for a total of 4 weeks. Of the 80 subjects, 8 with a central venous catheter, including an umbilical venous catheter, were allocated to the intensive PK sampling group, and the rest (72 subjects) were allocated to the sparse sampling group (see Fig. SA1 in the supplemental material). Serial blood samples (0.5 ml) were obtained to determine the plasma fluconazole concentration predose and 0.5, 1, 3, 10, 24, 48, and 72 h after the third dosing for the intensive sampling group. In the sparse sampling group, one predose and three randomly ordered extra blood samples of 0.5 ml from scheduled laboratory analyses were used from each infant during intensive care after the third dosing of fluconazole.

Determination of the plasma fluconazole concentration.

Plasma fluconazole concentrations were determined using a validated method, as follows. Briefly, the Agilent 1260 high-performance liquid chromatography (HPLC) system (Agilent, Palo Alto, CA, USA) was equipped with a Luna C₁₈ column (5 μm, 2.0 by 100 mm; Phenomenex, Torrance, CA, USA). This HPLC system was coupled with an Applied Biosystems API Qtrap 4000 tandem mass spectrometry (MS/MS) instrument (Applied Biosystems, Foster City, CA, USA). Quantification was conducted using an electrospray ionization (ESI) source in the positive mode and the following conditions: curtain gas of 20, gas 1 (nebulizer gas) of 50, gas 2 (heater gas) of 50, charged aerosol detection gas on high, TurboIonSpray (IS) voltage of 5,500 V, entrance potential (EP) of 10 V, collision energy (CE) of 25 V, probe temperature of 500°C, and scan time of 250 ms. HPLC separation was performed using an isocratic mobile phase composed of 10 mM ammonium formate in distilled water (0.1% formic acid) and 0.1% formic acid in acetonitrile (65:35, vol/vol) on a Luna guard column (2 μm, 4.0 by 2.0 mm; Phenomenex, Torrance, CA, USA). The flow rate was 0.3 ml/min, and the injection volume was 2 μl, with a 3-min run time. The multiple-reaction monitoring mode (MRM) was used for MS/MS detection at m/z 307.129 > 238.200 for fluconazole and at m/z 311.178 > 242.200 for fluconazole-d4 (internal standard). The range of quantification was 0.001 to 2 mg/liter. The accuracy (expressed as the percent difference from the theoretical concentration) of the quality control samples used during the sample analysis was in the range of 91.7 to 95.4%, with a precision (expressed as a coefficient of variation) of ≤5.2%.

Population pharmacokinetic analysis.

Plasma concentration-time records obtained from the eligible preterm infants were used for population PK analysis. The first-order conditional estimation method with the η-ε interaction option (FOCE-I) was used throughout the model-building process. Both single- and multicompartmental models were tested to describe the distribution of fluconazole PK. Different model parameterization scenarios were tested using the nonlinear mixed-effects method in NONMEM version 7.4.0 (Icon Development Solutions, Ellicott City, MD, USA) (57). Individual model parameters were obtained by a Bayesian estimation implemented in NONMEM with a covariance step. The interindividual variabilities of the PK parameters were tested in the model, along with various residual-error models for the intraindividual variabilities.

GA and postmenstrual ages (PMA) in weeks, postnatal age (PNA) in days, body weight (WT), height (HT), and the following baseline measures of laboratory tests were included in the covariate screening: alanine transaminase (ALT) and aspartate transaminase (AST) levels, serum creatinine level (CRE), estimated glomerular filtration rate (eGFR) (using the bedside Schwartz formula for children, 0.413 × HT/CRE), and blood urea nitrogen level (BUN) (41). For time-varying observations such as WT and CRE, more than one value per infant was included since the baseline observation was over the fluconazole treatment period. The covariates were screened based on a forward-selection significance level of 0.05 and a backward-deletion criterion of 0.01 for the objective function values (OFVs).

Successful minimization and standard basic model diagnostic plots, including individual fittings, goodness-of-fit (GOF) plots, and visual predictive check (VPC) plots for observed and predicted plasma fluconazole concentrations, were examined to assess the model fitting and the performance of the predictions. In addition, the precision of the parameter estimates was evaluated using 500 bootstrap-resampled simulated trials from the original data set that were sequentially estimated using the same final model for comparison of the medians and 95% confidence intervals (CIs) (2.5th and 97.5th percentiles) to those of the final model parameter estimates. The bootstrap procedures and VPC plot construction were performed with Perl-speaks-NONMEM (PsN) (version 4.7.0) (58).