First Dose in Neonates: Are Juvenile Mice, Adults and In Vitro–In Silico Data Predictive of Neonatal Pharmacokinetics of Fluconazole

Abstract

Background and objectives

Selection of the first-dose-in-neonates is challenging. The objective of this proof-of-concept study was to evaluate a pharmacokinetic bridging approach to predict a neonatal dosing regimen.

Methods

We selected fluconazole as a paradigm compound. We used data from studies in juvenile mice and adults to develop population pharmacokinetic models using NONMEM. We also develop a physiologically-based pharmacokinetic model from in vitro–in silico data using Simcyp. These three models were then used to predict neonatal pharmacokinetics and dosing regimens for fluconazole.

Results

From juvenile mice to neonates, a correction factor of maximum lifespan potential should be used for extrapolation, while a “renal factor” taking into account renal maturation was required for successful bridging based on adult and in vitro–in silico data. Simulations results demonstrated that the predicted drug exposure based on bridging approach was comparable to the observed value in neonates. The prediction errors were −2.2, +10.1 and −4.6 % for juvenile mice, adults and in vitro–in silico data, respectively.

Conclusion

A model-based bridging approach provided consistent predictions of fluconazole pharmacokinetic parameters in neonates and demonstrated the feasibility of this approach to justify the first-dose-in-neonates, based on all data available from different sources (including physiological informations, preclinical studies and adult data), allowing evidence-based decisions of neonatal dose rather than empiricism.

Appendix

1.1 Methods

We derived the population pharmacokinetic models from juvenile mice and adult healthy volunteers, and PBPK model from in vitro–in silico data to predict fluconazole dose in neonates.

1.1.1 Population Pharmacokinetic Parameters in Neonates

The data used to develop the population pharmacokinetic model was obtained from a prospective, single-center, open-label pharmacokinetic trial at Duke University Medical Center in Durham, NC. Thirteen neonates and young infants were given an intravenous loading dose (25 mg/kg administered over 2 h via a syringe pump) followed by maintenance therapy (12 mg/kg/day over 1 h every 24 h). Blood samples were taken at the following time points after the end of the infusion: 0–30 min, 2–4 h, 6–12 h, and 18–24 h after the first dose and peak and trough samples at doses 3 and 5 [32]. A summary of study characteristics and the demographic variables is provided in Table 1.

A total of 94 fluconazole concentrations ranging from 6.9 to 47.2 µg/mL were available. The population pharmacokinetic analysis was carried out using the nonlinear mixed effects modeling program NONMEM V 7.2 (Icon Development Solutions, Ellicot City, MD, USA). We used first order conditional estimation (FOCE) method with interaction option to estimate pharmacokinetic parameters and their variability. We estimated inter-individual variability of the pharmacokinetic parameters using an exponential model, which was expressed by Eq. (2):

$$ \theta_{i} = \, \theta_{\text{mean}} \times {\text{ e}}^{\eta i} $$

(2)

where θ _i represents the parameter value of the ith subject, θ _mean the typical value of the parameter in the population and ηi the variability between subjects which is assumed to follow a normal distribution with a mean of zero and variance ω ².

Covariate analysis followed a forward and backward selection process. We used stepwise covariate modelling [33] and likelihood ratio test to evaluate the effect of each variable. During the first step of covariate model building, a covariate was included if a significant (p < 0.05, χ ² distribution with one degree of freedom) decrease (reduction >3.84) in the objective function value (OFV) from the basic model was obtained. All the significant covariates were then added simultaneously into a ‘full’ model. Subsequently each covariate was independently removed from the full model. If the increase in the OFV was lower than 6.635 (p < 0.01, χ ² distribution), the covariate was considered significantly correlated to the pharmacokinetic parameter and was therefore included in the final model.

Model validation was based on graphical and statistical criteria. Goodness-of-fit plots, including observed concentration versus individual prediction, observed concentration versus population prediction (PRED), conditional weighted residuals (CWRES) versus time and CWRES versus PRED were used initially for diagnostic purposes [34]. The stability and performance of the final model were also assessed by means of a nonparametric bootstrap with re-sampling and replacement. Re-sampling was repeated 500 times and the values of estimated parameters from the bootstrap procedure were compared with those estimated from the original data set. The entire procedure was performed in an automated fashion, using PsN (v2.30) [35]. The final model was also evaluated graphically and statistically by normalized prediction distribution errors (NPDE) [36]. 1,000 datasets were simulated using the final population model parameters. NPDE results were summarised graphically by default as provided by the NPDE R package (v1.2) [37]: (i) QQ-plot of the NPDE; (ii) histogram of the NPDE. The NPDE is expected to follow the N (0, 1) distribution.

The predictive performance of the developed model was further evaluated in an independent group of neonates. The data was obtained from a prospective, single-center, open-label pharmacokinetic trial at the neonatal intensive care unit of Hospital for Children and Adolescents, University of Helsinki. Twelve preterm neonates were given fluconazole (6 mg/kg administered over 15 min via an infusion pump every 72 h) for prophylaxis therapy. Blood samples were collected before and at the end of the infusion, as well as at 3, 6, 24, 48, and 72 h after the first infusion was started. The same sampling was repeated after the third and fifth doses [38].

1.1.2 From Juvenile Mice to Neonates

The pharmacokinetic data in juvenile mice (Swiss strain) was obtained from an open-label, repeated-doses pharmacokinetic study within the FP7 TINN project (Treatment Infection in NeoNates). The mice were born at 19 days gestation age. Two dosing regimens 6 or 60 mg/kg once daily were given subcutaneously since postnatal day 2 (Table 1). Pharmacokinetic samplings were obtained at times of T0.5 h or T23 h on day 1, T23 h on day 5, T23 h on day 10, or T0.5 h or T23 h on day 11.

Population pharmacokinetic analysis was carried out using NONMEM. Modeling and validation processes were similar to neonates, as described above.

We used the following extrapolation approaches to predict pharmacokinetic parameters in neonates from juvenile mice [22, 39–41]:

1.1.2.1 Clearance (CL)

Method 1: Weight normalized CL [Eq. (3)]:

$${\text{CL}}_{\text{NN}} = a \times {\text{WT}}_{\text{NN}} $$

(3)

where CL_NN is the predicted CL in neonates, WT_NN is the bodyweight in neonates; a is the estimated weight normalized CL in juvenile mice.

Method 2: Simple allometry [Eq. (4)]:

$$ {\text{CL}}_{\text{NN}} = b \times {\text{WT}}_{\text{NN}}^{c} $$

(4)

where CL_NN is the predicted CL in neonates, WT_NN is the bodyweight in neonates; b and c are the estimated coefficient and exponent obtained from the population pharmacokinetic model derived from juvenile mice.

Method 3: Simple allometry with corrected factor of maximum lifespan potential [Eqs. (5 and 6)]:

$$ {\text{CL}}_{\text{NN}} = \frac{{d \times {\text{WT}}_{\text{NN}}^{e} }}{81,800} $$

(5)

$$ {\text{MLP (years)}} = 185.4 \times ( {\text{Brain weight, kg)}}^{0.636} \times ( {\text{Body weight, kg)}}^{ - 0.225} $$

(6)

Maximum lifespan potential (MLP) is an estimate of the maximum amount of time that a member of a given species could survive. It has been used to correct the prediction bias of interspecies scaling from animal to human [39]. In this approach, the individual Bayesian estimated CL in mice was multiplied by its MLP value and were plotted as a function of bodyweight on a log–log scale. The coefficient and exponent (d and e in the equation) were estimated from the allometric equation. The MLP was calculated by the equation as described by Sacher [42]. Brain weight values are 120 mg on day 1, 248 mg on day 5 and 314 mg on day 11 [43], whereas 81,8000 h is the MLP in humans.

1.1.2.2 Volume of Distribution (V _d)

Method 1: Weight normalized V _d [Eq. (7)]:

$$ V_{{{\text{d}}_{\text{NN}} }} = f \times {\text{WT}}_{\text{NN}} $$

(7)

where Vd_NN is the predicted Vd in neonates, WT_NN is the bodyweight in neonates; f is the estimated weight normalized Vd in juvenile mice.

Method 2: Simple allometry [Eq. (8)]:

$$ Vd_{NN} = g \times WT_{NN}^{h} $$

(8)

where $V_{{{\text{d}}_{\text{NN}} }}$ is the predicted V _d in neonates; WT_NN is the bodyweight in neonates; g and h are the estimated coefficient and exponent obtained from the population pharmacokinetic model of juvenile mice.

Although the bioavailability (F) of fluconazole is unknown in juvenile mice whatever the route of administration, “reference values” do exist showing that fluconazole is almost completely absorbed both in animals and humans: F was 108 % in sea turtles after subcutaneous administration, >90 % in human and 109 % in cats after oral administration [44, 45]. Thus, we assumed a value of 100 % for F after sub-cutaneous administration in juvenile mice and oral administration in human.

1.1.3 From Adult Health Volunteers to Neonates

The pharmacokinetic data in adult healthy volunteers was identified from literature [46]. In a phase 1 study, 12 healthy male volunteers received a single oral dose of fluconazole 100 mg. An average of 15 blood samples per patient was taken between 5 min and 168 h postdose. A one-compartment population pharmacokinetic model was developed based on data from this pharmacokinetic study [14], which was used in the present study.

We used the following extrapolation approach to predict pharmacokinetics in neonates from adult healthy volunteers [23, 25, 47]:

Method 1: Fixed-exponent allometry with corrected factor of postmenstrual age for CL [Eqs. (9, 10 and 11)]

$$ {\text{CL}}_{\text{NN}} = {\text{CL}}_{\text{adult}} \times \left( {\frac{{{\text{WT}}_{\text{NN}} }}{70}} \right)^{0.75} \times\, {\text{MF}} $$

(9)

$$ {\text{MF}} = \frac{{{\text{PMA}}^{S} }}{{{\text{PMA}}_{50}^{S} + {\text{PMA}}^{S} }} $$

(10)

$$ V_{{{\text{d}}_{\text{NN}} }} = V_{{{\text{d}}_{\text{adult}} }} \times \left( {\frac{{{\text{WT}}_{\text{NN}} }}{70}} \right)^{1} $$

(11)

where CL_NN and $V_{{{\text{d}}_{\text{NN}} }}$ are the predicted CL and V _d in neonates; CL_adult and $V_{{{\text{d}}_{\text{adult}} }}$ are the observed CL and V _d in adults, respectively; WT_NN is the bodyweight in neonates. MF is maturation function, PMA is postmenstrual age in weeks, PMA₅₀ is PMA at which CL reaches half its maximal value and S is sigmoidicity coefficient. To describe renal maturation, PMA₅₀ and S were set to 47.7 weeks and 3.4, as previously reported [25].

Method 2: Fixed-exponent allometry of 0.75 for CL [Eq. (12)]

$$ {\text{CL}}_{\text{NN}} = {\text{CL}}_{\text{adult}} \times \left( {\frac{{{\text{WT}}_{\text{NN}} }}{70}} \right)^{0.75} $$

(12)

Method 3: Age dependent exponent allometry for CL [Eq. (13)]

$$ {\text{CL}}_{\text{NN}} = {\text{CL}}_{\text{adult}} \times \left( {\frac{{{\text{WT}}_{\text{NN}} }}{70}} \right)^{{{ \exp }\_{\text{age}}}} $$

(13)

where exp_age is age dependent exponent: 1.2 for children ≤3 months and 1.0 for children >3 months to 1 year.

1.1.4 From In Vitro–In Silico Data to Neonates

The PBPK model was carried out using the software Simcyp version 12 release 1 (Simcyp Limited, Sheffield, UK). Drug data (e.g., drug molecular weight, physico-chemical characteristics such as logP (octanol–water partition coefficient) and pKa, drug elimination etc.) was obtained from the literature. The Systems data in children (e.g. physiology, anatomy, biology, and biochemistry defined based on pediatric population demographics) was set to default values, as described by Johnson et al [28].

Extrapolation approach

The default kidney maturation model in Simcyp [Eq. (14)] was used for extrapolation [28]:

$$ {\text{Glomerular filtration rate (mL/min}}) = ( - 6. 1 60 4 \times {\text{Body surface area}}^{ 2} ) + ( 9 9.0 5 4 \times {\text{Body surface area}}) - 1 7. 7 4 $$

(14)

1.1.5 Evaluation of Model-Based Extrapolations

Given our interest in dosage prediction using model-based approaches, we tested the performance of the proposed approaches via simulation. We extensively evaluated whether the two population pharmacokinetic models derived from juvenile mice and adults, and the PBPK model derived from in vitro–in silico data with associated extrapolation approaches could be used to accurately predict observed pharmacokinetic parameters in neonates. We elected CL and V _d as endpoints for the purpose of this evaluation and performed 100 simulations using NONMEM and Simcyp, respectively.

1.1.6 Predictive Performance of the Model in Dosage Prediction

Fluconazole pharmacokinetic–pharmacodynamic relationship was best described by using the ratio of area under curve and minimal inhibition concentration (AUC/MIC). The selection of pediatric dose is based on the target exposure (AUC_0–24) of 800 mg·h/L, reported in adult intensive care patients or immunocompromised patients treated with fluconazole at 800 mg/day. This AUC target ensures that exposure exceeds the AUC/MIC pharmacodynamic target of 50 h for Candida species with a MIC of 8 mg/L [12, 48, 49].

In the next step, we tested different simulation scenarios to evaluate whether model-based extrapolation could be used to accurately predict drug exposure and support the dosage prediction in neonates. The loading dose regimen previously recommended for neonates [32], consisting of a loading dose of 25 mg/kg (intravenous infusion over 2 h) followed by maintenance doses of 12 mg/kg (intravenous infusion over 1 h) was used to perform 100 simulations using our models previously developed: population pharmacokinetic models in mice and adults, PBPK model and population pharmacokinetic model in neonates to calculate predicted and reference AUCs, respectively.