From PICU to NICU: Extrapolating Meropenem Exposure From Pediatric to Neonatal Intensive Care Patients

Abstract

We previously developed a population pharmacokinetic (PopPK) model for meropenem in pediatric intensive care unit patients accounting for effect of body size, maturation, and kidney function on clearance. This study aimed to extrapolate meropenem exposure to neonates and young infants using the pediatric PopPK model and to validate the predictions using external data. An independent dataset was obtained from the regulations.gov website, which included 176 neonates and young infants (up to 3 months old) with a total of 767 plasma meropenem concentrations. After normalizing the estimated glomerular filtration rate (eGFR) using a maturation factor, the PopPK model was applied to this dataset and the concordance between the model predictions and observed concentrations was visually assessed using goodness‐of‐fit (GOF) plots and prediction‐corrected visual predictive check. Median prediction error (MDPE) evaluated bias and median absolute prediction error (MDAPE) evaluated precision of the predictions. GOF plots indicated no apparent bias or model misspecification. Individual‐level predictions showed an MDPE of 1% and an MDAPE of 18.3%, both within commonly accepted thresholds for bias (<±20%‐30%) and precision (<30%‐35%), respectively. The findings support the model's application for simulations when neonatal eGFR is normalized using a maturation factor and for model‐informed precision dosing in clinical practice for neonates and infants.

Introduction

Developing new drugs and accurately determining appropriate dosing in children and neonates remain major challenges due to underfunding, ethical constraints, and complex regulatory requirements. ¹ Drug development typically follows a sequential approach, beginning with studies in adults, followed by studies in older children, and only later extending to neonates. This process creates a substantial opportunity to develop and refine methodologies for predicting the pharmacokinetics of specific drugs in neonates using data from older populations, while accounting for physiological differences in early life. ²

In pediatrics, drug clearance is governed by a complex interplay of growth, developmental maturation, and organ function. ³ It has been proposed that population pharmacokinetic (PopPK) models in children and neonates can be described with allometric scaling of body weight to account for growth (F_size), along with a sigmoidal function of postmenstrual age (PMA) to capture age‐dependent maturation (F_maturation), which reflects a fraction of adult clearance. ⁴ Rhodin et al used a sigmoidal maturation function to explain the increase in glomerular filtration rate (GFR) with age, ⁵ while other studies have applied it to explain the maturation of drug clearance. ⁶ In the case of primarily renally eliminated drugs, organ function is mainly represented by kidney function (F_kidney), which is frequently estimated using biomarkers such as serum creatinine or estimated GFR (eGFR). ⁷

Although these are interdependent factors, when the three functions —F_size, F_maturation, and F_kidney—are incorporated into a PopPK model, F_maturation represents maturation independent of body size, while F_kidney reflects only deviations from normal kidney function. ³ It is important to recognize that eGFR values reflecting normal kidney function in neonates are significantly lower than those in older patients, ⁸ and these differences must be accounted for in a model. In this study, we demonstrate how a PopPK model developed for older patients can successfully predict drug concentrations in neonates for a β‐lactam antibiotic when accounting for these three key factors influencing drug clearance.

Meropenem is a broad‐spectrum β‐lactam antibiotic commonly used to treat severe or drug‐resistant bacterial infections. ⁹ In our previous work, we developed a meropenem PopPK model using data from pediatric intensive care unit (PICU) patients aged 4 months to 30 years, with the intention of applying it for model‐informed precision dosing (MIPD). ¹⁰ However, it is essential to evaluate the model's predictive performance of meropenem exposure in neonates and young infants before it can be used for MIPD in the neonatal intensive care unit (NICU). ¹¹

Therefore, the aim of this study was to assess the extrapolation of an existing meropenem PopPK model, developed in older pediatric patients and young adults, to neonates and young infants using an independent external dataset.

Material and Methods

Meropenem Model Description

A PopPK model of meropenem was previously developed using real‐world data from critically ill pediatric and young adult patients. ¹⁰ The model‐building dataset was obtained from a study approved by the Cincinnati Children's Hospital Medical Center Institutional Review Board, which granted a waiver of consent (IRB #2018‐3245). Briefly, it is a two‐compartment model that integrates body weight allometric scaling (F_size), cumulative percentage of fluid balance as a covariate on the central volume of distribution (V1), eGFR (F_kidney) as a covariate on clearance (CL), and a fixed maturation factor (F_maturation) on CL. This model describes individual values for CL, V1, intercompartmental clearance (Q), and peripheral volume of distribution (V2) as follows:

$$\begin{array}{ccc}\hfill \mathrm{C}Li& =& 13.22\times {\left(\frac{WT}{70}\right)}^{0.75}\times {\left(\frac{\mathrm{e}GFR}{140}\right)}^{0.45}\hfill \\ & & \times \frac{{\mathrm{P}MA}^{Hill}}{{\mathrm{P}MA}^{Hill}+\mathrm{P}MA50^{Hill}}\times {\mathrm{e}}^{\eta IIV+\eta \mathrm{I}OV}\hfill \end{array}$$ (1)

$$\mathrm{V}1i=16.45\times \left(\frac{\mathrm{W}T}{70}\right)\times {\mathrm{e}}^{0.033\times \mathrm{C}um\%\mathrm{F}B}\times {\mathrm{e}}^{\eta \mathrm{I}IV+\eta \mathrm{I}OV}$$ (2)

$$\mathrm{Q}i=1.78\times {\left(\frac{\mathrm{W}T}{70}\right)}^{0.75}$$ (3)

$$\mathrm{V}2i=7.66\times \left(\frac{\mathrm{W}T}{70}\right)$$ (4)

ETA (η) represents the random effects that account for both interindividual variability (IIV) and interoccasion variability (IOV). The maturation factor follows a sigmoid curve, where PMA represents PMA in weeks, the Hill coefficient characterizes the steepness of the maturation curve, and PMA₅₀ represents the age at which drug clearance reaches 50% of the adult values. As meropenem is primarily eliminated by glomerular filtration, this model assumes that the maturation of meropenem clearance follows the same pattern as the GFR maturation curve described by Rhodin et al, who reported that PMA₅₀ is 47.7 weeks and the Hill coefficient is 3.4 for GFR maturation from birth to adult levels. ⁵

Neonates And Infants Data

Meropenem concentrations and clinical data from 176 hospitalized neonates and infants were retrieved from the regulations.gov website, a repository for clinical data submitted to regulatory authorities. ¹² The original clinical trial was approved by institutional review boards at participating sites and registered at ClinicalTrials.gov (NCT00621192). Written informed consent was obtained from the parent or legal guardian of each participant prior to enrollment. In the original study, meropenem was given at 20‐30 mg/kg every 8‐12 h based on clinical indications, with timed and opportunistic sampling used for concentration measurement. A total of 767 meropenem concentrations were included in the analysis. The detailed study protocol was previously described. ¹²

The study population had PMA ranging from 24 to 44 weeks, covering a spectrum from extremely preterm to full‐term newborns, and postnatal ages from 1 to 90 days. Specifically, the cohort included 90 extremely preterm neonates (gestational age <28 weeks), 35 very preterm (28 to <32 weeks), 32 moderate to late preterm (32 to <37 weeks), and 19 term neonates (≥37 weeks). The external dataset included comprehensive clinical and demographic information, including age, weight, height, sex assigned at birth, serum creatinine, albumin, urine output, and use of vasoactive drugs. However, it lacked fluid balance data, a covariate of our meropenem PopPK model; thus, fluid balance was assumed to be zero for all patients.

We compared characteristics of patients in the model‐building cohort and the neonates and infants in the external dataset using the Mann–Whitney U test for continuous variables and the Chi‐square test for categorical variables. A P‐value <  .05 was considered significant.

Estimation of Kidney Function

GFR was estimated using the bedside Schwartz equation. ¹³ A coefficient (k) of 0.31 was applied for neonates younger than 1 month, as proposed by Smeets et al, ¹⁴ while the original coefficient of 0.413 was used for patients older than 1 month. Although the estimated GFR is standardized to a typical adult body surface area (BSA of 1.73 m²), it does not account for age‐related kidney maturation. Neonatal eGFR values are substantially lower than those of older children and adults, even after BSA standardization, due to the developmental immaturity of kidney function. Therefore, we normalized neonatal eGFR values to adult‐equivalent levels by dividing by a GFR maturation factor, using values reported by Rhodin et al (PMA₅₀ = 47.7 weeks, Hill coefficient = 3.4): ⁵

$$\begin{array}{c}\hfill \mathrm{Adult}-\mathrm{normalized}\mathrm{eGFR}=\\ \hfill \frac{\mathrm{Neonate}\mathrm{eGFR}}{\mathrm{GFR}\mathrm{Maturation}\mathrm{Factor}}\end{array}$$ (5)

$$\mathrm{GFR}\mathrm{Maturation}\mathrm{Factor}=\frac{{\mathrm{P}MA}^{3.4}}{\mathrm{PM}{\mathrm{A}}^{3.4}+{47.7}^{3.4}}$$ (6)

This scaling ensures that the eGFR term reflects only the deviation from normal kidney function (F_kidney), independent of the infant's body size or age, as these effects are already accounted for separately through allometric scaling (F_size) and a maturation function (F_maturation) in the model.

Assessing the PopPK Model Performance in Neonates And Infants

The PopPK model was used to predict meropenem concentrations in neonates and infants using Monolix (2024R1, Lixoft, France). Prediction errors (PE) were calculated as the difference between predicted and observed concentrations in the external dataset, expressed as a percentage relative to the observed value. Median prediction error (MDPE) and median absolute prediction error (MDAPE) were calculated to quantify, respectively, the bias and precision of the population and individual predictions. Goodness‐of‐fit (GOF) plots and a prediction‐corrected visual predictive check (pcVPC) were generated to assess the model's predictive performance.

Results

We compared the clinical and demographic data of our model‐building cohort with the external dataset cohort (Table 1). The median age and body weight in the neonate and infant cohort were significantly lower, reflecting the different age groups in each study. Initial eGFR value was also significantly different in the neonates and infants cohort, due to immature kidney function in younger patients. However, after normalizing eGFR values to adult‐equivalent levels using the GFR maturation factor, the difference between cohorts was no longer statistically significant (P >  .05), indicating that both study cohorts have comparable kidney function status after accounting for age and body size differences.

Table 1.

Demographics and Clinical Characteristics of Patients Receiving Meropenem in the Model‐Building and External Validation Cohorts

Variable	Model‐Building Cohort (n = 48)1	External Validation Cohort (n = 176)2	P‐value
Demographic data
Age (range)	13.4 years (4 months to 30 years)	21.5 days (1 day to 3 months)	<.05*
Gestational age (weeks), median (IQR)	34.5 (26.5‐38.8)	27.5 (25‐33)	.25
Neonates (<1 month), n (%)	0 (0%)	112 (64%)	<.05*
≤7 days old	‐	21 (12%)	‐
>7 to ≤14 days old	‐	42 (24%)	‐
>14 to ≤30 days old	‐	49 (28%)	‐
Infants (1 month to 2 years), n (%)	8 (17%)	64 (36%)	<.05*
Children (2 to 12 years), n (%)	14 (29%)	0 (0%)	<.05*
Adolescents (12 to 17 years), n (%)	13 (27%)	0 (0%)	<.05*
Young adults (17 to 30 years), n (%)	13 (27%)	0 (0%)	<.05*
Male, n (%)	30 (62%)	102 (58%)	.57
Admission body weight (kg), median (IQR)	29 (16.2‐55.7)	1.5 (1.1‐2.3)	<.05*
Height (cm), median (IQR)	128.8 (97.9‐162.4)	39.5 (34.8‐45.0)	<.05*
Clinical data
Serum creatinine (mg/dL), median (IQR)	0.4 (0.2‐0.6)	0.5 (0.3‐0.8)	<.05*
eGFR (mL/min/1.73 m2), median (IQR)	133 (94.1‐170.8)	26.9 (15.9‐50.9)	<.05*
Adult‐normalized eGFR (mL/min/1.73 m2), median (IQR)	133 (94.1‐173.2)	128.9 (90.1‐196.6)	.75
Cumulative % fluid balance
Day 1 (n=48), median (IQR)	3 (1.3‐5.8)	NA	‐
Day 7 (n=23), median (IQR)	10.9 (7.3‐21.7)	NA	‐

Adult‐normalized eGFR: Neonatal eGFR values were normalized to adult‐equivalent levels by dividing by the GFR maturation factor (PMA₅₀ = 47.7 weeks, Hill coefficient = 3.4); eGFR, estimated glomerular filtration rate calculated using the Schwartz equation with coefficient (k) of 0.31 for neonates (<1 month old) and 0.413 for older patients; IQR, interquartile range; NA, not available.

Statistically significant values, based on the Mann–Whitney U test for continuous variables and the Chi‐square test for categorical variables, are indicated with an asterisk (*) in the table, denoting a P‐value <  .05.

¹ Morales Junior et al. (2024).

² Smith et al. (2011).

PopPK Model Performance in Neonates and Infants

The evaluation of extrapolation performance using our PopPK model to predict meropenem concentrations in the neonatal and infant cohort yielded a population MDPE of 5.3% and a population MDAPE of 36.0%. For individual predictions, the MDPE was 1.0%, while the MDAPE stood at 18.3%.

The GOF plots (Figure 1) did not display any systematic deviations. In the observations versus predictions plots, the data points are symmetrically distributed around the identity line. The pcVPC showed that the observed concentrations mostly fell within the 95% confidence interval of the model‐predicted intervals (Figure 2).

Figure 1.

Population (left) and individual (right) model predictions versus observations for meropenem concentrations (mg/L). The solid line represents the line of unity (y = x).

Figure 2.

Prediction‐corrected visual predictive check (pcVPC) for meropenem concentrations. Black dots represent observed concentrations in neonates and young infants. Solid lines represent the median, 5th, and 95th percentile of the observed data, while shaded areas show the 95% prediction intervals (n = 1000) generated from the original population pharmacokinetic model developed in older children and young adults.

A comparison of the GOF plots (Figure S1) and error metrics (Table S1) before and after GFR normalization is presented in the Supplemental Information. This comparison shows that the normalization of eGFR values to adult‐equivalent values using a maturation factor improved alignment between observed and predicted concentrations, as failing to normalize eGFR significantly underpredicts meropenem clearance in the younger cohort.

Figure 3 illustrates the individual meropenem clearance of the neonates and infants, normalized to a 70 kg body weight and adult‐normalized eGFR of 140 mL/min/1.73 m², expressed as a percentage of typical adult meropenem clearance. This figure demonstrates that our model effectively predicts meropenem clearance in neonates and infants by incorporating age, body weight, and kidney function.

Figure 3.

Individual meropenem clearance adjusted to 70 kg body weight and adult‐normalized estimated glomerular filtration rate (GFR) of 140 mL/min/1.73 m² expressed as a percentage of typical adult clearance, plotted against postmenstrual age (PMA). The shaded area indicates the 95% confidence interval (CI) of meropenem clearance predicted by the PopPK model, while the solid black lines represent the median, 2.5th, and 97.5th percentiles of the predicted clearance. Circles represent individual meropenem clearance values from patients in the study by Smith et al.,¹² while black diamonds represent individual meropenem clearance values from patients under the age of 4 years from the model‐building cohort described by Morales Junior et al.¹⁰

Discussion

In this study, we successfully extrapolated a previously developed meropenem PopPK model, originally established for PICU patients, to neonates and infants.

The inclusion of functions for body size, maturation, and kidney function, along with the innovative approach of normalizing eGFR to an adult‐equivalent value, enabled the application of the meropenem PopPK model across different age groups. Germovsek et al developed a one‐compartment PopPK model of meropenem based on data from neonates and infants, which also incorporated these three functions for clearance. ¹⁵ A notable distinction from our approach lies in their use of serum creatinine concentration (standardized by PMA to adjust for maternal creatinine contribution and age‐related renal maturation) as a marker of kidney function. Similarly, Lonsdale et al developed models describing the age‐related maturation of β‐lactam antibiotics, also incorporating serum creatinine as a renal biomarker. ¹⁶ Several other PopPK models for renally eliminated antibiotics in neonates have followed this approach, applying a sigmoidal maturation function for drug clearance and including serum creatinine, instead of eGFR, to reflect kidney function. ⁴ , ¹⁷ , ¹⁸

eGFR is widely used both in clinical practice and pharmacokinetic research, as dosing adjustments in FDA‐approved drug labels are based on eGFR rather than serum creatinine. In our study, we used eGFR as a marker of kidney function, normalizing it to adult‐equivalent values using an established GFR maturation factor based on PMA. ⁵ This approach provides a physiologically sound and scalable method to predict the impact of kidney function on drug clearance across diverse pediatric age groups, independent of body size, and maturation effects. A similar methodology was employed by Holford et al, who modeled the clearance of gentamicin, amikacin, and vancomycin in neonates by incorporating allometric scaling and maturation functions, as well as a term that represents GFR for a typical adult male without kidney disease (70 kg, 176 cm), but with a different parameterization based on both PMA and postnatal age. ¹⁹ Since our cohort included a limited number of patients in the immediate postnatal period, with only 12% being 7 days old or younger, we opted to use the Rhodin model based on PMA. ⁵ Using a different approach, De Cock et al described the developmental changes in the clearance of gentamicin, tobramycin, and vancomycin from preterm neonates to adults using a semi‐physiological model that characterized GFR maturation with a bodyweight‐dependent allometric exponent. ²⁰ Alternative modeling strategies have also been explored. Previous work on milrinone from our group demonstrated that drug clearance in infants and neonates can be effectively characterized by combining allometric scaling, maturation factors, and acute kidney injury (AKI) stages. ²¹ In that study, AKI was used as a specific marker of kidney function, enabling us to delineate its distinct impact alongside age‐dependent renal maturation. These modeling strategies hold promise in addressing the complex challenge of exposure extrapolation and dosing pediatric patients with varying kidney function. ²²

Most newly developed antibiotics targeting multidrug‐resistant bacteria are either novel β‐lactams or β‐lactam–β‐lactamase inhibitor combinations. ²³ These agents remain a cornerstone in treating infections caused by both Gram‐positive and Gram‐negative bacteria, including multidrug‐resistant organisms. Approximately 80% of the PopPK models developed for β‐lactam antibiotics include estimated or measured GFR as a significant covariate, typically normalized to a reference adult GFR or the median GFR of the study population. ²⁴ Since most of these agents are first approved or studied in adults, a potential strategy to support pediatric dosing is to extrapolate PopPK models developed in older patients considering the differences in body size, maturation, and kidney function, rather than relying solely on body weight‐based scaling. ²⁵

While there is no consensus on threshold error metrics for the external evaluation of PopPK models, previous studies have established 20%‐30% for bias and 30%‐35% for precision. ²⁶ For population‐level predictions, the model's MDPE was 5.3%, within the accepted bias range. The population MDAPE was 36.0%, slightly above the typical precision threshold, likely due to missing fluid balance data, which is a covariate in the meropenem PopPK model that affects the volume of distribution. ²⁷ In neonates and infants, small variations in fluid retention or hydration can significantly impact volume of distribution. By not accounting for these fluctuations, the model's precision at the population level was reduced. Nevertheless, the model's strong performance at the individual level (1.0% MDPE, 18.3% MDAPE) underscores its potential reliability and accuracy for MIPD in clinical practice.

This study has some limitations. The absence of fluid balance data in the external dataset likely contributed to the slightly reduced precision in population‐level predictions, as discussed earlier. Additionally, the dataset was obtained from a regulatory source rather than a prospective clinical study and included a mix of timed and opportunistic samples, potentially introducing variability in sampling times and clinical data quality, which likely contributed to the outliers observed in concentration predictions. Another limitation is the estimation of eGFR using the bedside Schwartz equation with a coefficient of 0.31, as proposed by Smeets et al for term‐born neonates. Since the majority of our cohort was preterm, this may have introduced inaccuracies in eGFR estimation for this population. Despite these limitations, the study demonstrates a practical approach for extrapolating PopPK models of renally eliminated drugs across age groups. Notably, the original model was developed without any neonatal data, yet it accurately predicted meropenem concentrations in neonates by applying pharmacology concepts. The diversity in demographics, clinical settings, and age ranges between the model‐building and external cohorts supports the model's generalizability. Strong individual‐level predictive performance (MDPE 1.0%, MDAPE 18.3%) further highlights its potential for MIPD in neonatal care, in addition to use in older pediatric and adult patients.

Conclusions

This study demonstrates that a meropenem PopPK model developed in older pediatric and adult ICU patients can successfully predict concentrations in neonates and young infants by accounting for body size, age‐related maturation, and kidney function. We employed an innovative approach by normalizing eGFR using a maturation factor, enabling the use of a single model to describe the changes in drug clearance throughout childhood. These findings support the model's application for simulations and MIPD in clinical practice to optimize meropenem therapy in neonatal and pediatric settings.

Author Contributions

Ronaldo Morales Junior designed the research, performed the research, analyzed the data, and wrote the manuscript. Wen Rui Tan, Kei Irie, Tomoyuki Mizuno, and Sonya Tang Girdwood designed the research, analyzed the data, and wrote the manuscript.

Conflicts of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

This work was generously supported by the National Institutes of Health (NIH), including funding from the National Institute of General Medical Sciences (NIGMS) under an R35 award (R35GM146701).

Data Sharing

The data that support the findings of this study are openly available and were obtained from Regulations.gov.

Supporting information

Supporting Information