Predictive Performance of a Gentamicin Pharmacokinetic Model in Term Neonates with Perinatal Asphyxia Undergoing Controlled Therapeutic Hypothermia

Abstract

Background: Model validation procedures are crucial when population pharmacokinetic (PK) models are used to develop dosing algorithms and to perform model-informed precision dosing. We have previously published a population PK model describing the PK of gentamicin in term neonates with perinatal asphyxia during controlled therapeutic hypothermia (TH), which showed altered gentamicin clearance during the hypothermic phase dependent on gestational age and weight. In this study, the predictive performance and generalizability of this model were assessed using an independent data set of neonates with perinatal asphyxia undergoing controlled TH.

Methods: The external data set contained a subset of neonates included in the prospective observational multicenter PharmaCool Study. Predictive performance was assessed by visually inspecting observed-versus-predicted concentration plots and calculating bias and precision. In addition, simulation-based diagnostics, model refitting, and bootstrap analyses were performed.

Results: The external data set included 323 gentamicin concentrations of 39 neonates. Both the model-building and external data set included neonates from multiple centers. The original gentamicin PK model predicted the observed gentamicin concentrations with adequate accuracy and precision during all phases of controlled TH. Model appropriateness was confirmed with prediction-corrected visual predictive checks and normalized prediction distribution error analyses. Model refitting to the merged data set (n = 86 neonates with 935 samples) showed accurate estimation of PK parameters.

Conclusions: The results of this external validation study justify the generalizability of the gentamicin dosing recommendations made in the original study for neonates with perinatal asphyxia undergoing controlled TH (5 mg/kg every 36 or 24 h with gestational age 36-41 and 42 wk, respectively) and its applicability in model-informed precision dosing.

FIGURE 1.

Predicted-versus-observed plots depicting gentamicin concentrations for all phases, including low levels (≤1.5 mg/L) and high levels (<2 hours after the previous dose) during controlled TH. Separated plots for the hypothermic and normothermic phases are provided in Supplemental Digital Content 1 (see Figure, http://links.lww.com/TDM/A705).

FIGURE 2.

Gentamicin pharmacokinetic model prediction error during all phases of controlled TH, encompassing the hypothermic, normothermic, and combined phases. Prediction error is assessed through bias, represented by (A) mean prediction error in mg/L, and precision, illustrated by (B) RMSE in mg/L. Closed circles denote the mean values with accompanying 95% confidence intervals. Low levels indicate gentamicin concentrations ≤1.5 mg/L, and high levels represent gentamicin concentrations between 0 and 2 hours after dosing. Prediction error for high levels during the normothermic phase is not presented because of the limited number of samples during this phase.

FIGURE 3.

Results obtained from the NPDE analysis, using (A) the model-building data set and (B) the external data set. The NPDE distribution displays a mean and variance of −0.0965 and 1.156, respectively, for the model-building data set, and −0.2087* and 1.726*, respectively, for the external data set (* indicates a statistically significant difference from 0 for mean and 1 for variance (P < 0.05) as determined by the Wilcoxon signed-rank test and Fisher test of variance). The TH period encompasses the hypothermic phase (0–72 hours), rewarming phase (72–96 hours), and normothermic phase (>96 hours).

FIGURE 4.

Prediction-corrected visual predictive checks (pcVPC) for (A) the external data set using the previously published gentamicin model and (B) the merged data set, which includes both the model-building and external data sets. The black open circles represent observed gentamicin concentrations, while the black solid line denotes the observed median, and the black dashed lines depict the 5th and 95th percentiles. The orange area signifies the 95% CI of the model-predicted median, and the blue area indicates the model-predicted 5th and 95th percentiles.