Clinical researchers analyzing therapeutic drug monitoring data have developed a population pharmacokinetic model demonstrating that body weight and estimated glomerular filtration rate serve as the primary determinants of vancomycin clearance in pediatric patients. The retrospective investigation, conducted at the Pediatric Department of Anhui Children’s Hospital, evaluated steady-state serum concentrations to address the clinical challenge of tailoring dosing regimens in children. By applying non-linear mixed-effects modeling to clinical observations collected between September 2021 and November 2023, the study established typical values for drug clearance and volume of distribution while quantifying how physiological differences alter systemic drug exposure in younger populations.
Vancomycin provides potent antibacterial activity against infections caused by methicillin-resistant Staphylococcus aureus and other Gram-positive pathogens, yet clinical management remains complicated by substantial pharmacokinetic variability. While the antimicrobial agent has been utilized in healthcare settings for more than six decades, dosing strategies for pediatric patients have historically relied on evidence derived from adult trials. Children possess different body surface areas, distinct extracellular fluid volumes, and immature renal filtration pathways relative to adults, which drives wider inter-individual variation in drug handling. Furthermore, vancomycin exhibits a narrow therapeutic window, with pediatric nephrotoxicity rates typically ranging from 5 to 20 percent and rising to between 30 and 40 percent when trough concentrations exceed 15 milligrams per liter.
To establish the empirical dataset, the investigators screened pediatric patients under 18 years of age who received intermittent intravenous vancomycin infusions for confirmed or suspected Gram-positive bacterial infections. Eligible subjects were required to have at least one documented steady-state trough concentration measurement. Patients who underwent renal replacement therapy or extracorporeal membrane oxygenation were excluded, along with those lacking complete demographic or pathophysiological records or presenting undetectable drug concentrations without numerical reporting. Samples displaying measurable concentrations below the lower quantification threshold of 2 milligrams per liter were retained for mathematical modeling, yielding a final analytical cohort of 100 pediatric patients and 124 measured serum concentrations.
Administration of vancomycin was performed using vials dissolved in 0.9 percent saline or 5 percent glucose delivered through intravenous infusions lasting 60 minutes. Venous blood samples of approximately 2 milliliters were collected in ethylenediaminetetraacetic acid tubes 30 minutes prior to the fourth scheduled dose, aligned with clinical recommendations from the Expert Consensus on Therapeutic Drug Monitoring in Children. Following centrifugation at 3500 revolutions per minute for 10 minutes, plasma supernatants of at least 0.3 milliliters were separated and analyzed via enzyme amplification immunoassay on an automated biochemical analyzer. Serum creatinine concentrations were standardized to micromoles per liter, and individual estimated glomerular filtration rates were determined using the pediatric Schwartz equation incorporating height and age-specific coefficients.
Pharmacokinetic parameter estimation was executed using non-linear mixed-effects modeling in NONMEM version 7.4.1 supported by Perl-speaks-NONMEM and R. Because the clinical dataset was derived from routine hospital monitoring, blood draws were sparsely distributed and dominated by pre-dose trough levels. The researchers adopted a one-compartment structural model with first-order elimination using the ADVAN1 TRANS2 module and first-order conditional estimation with eta-epsilon interaction. Inter-individual parameter variability was represented through exponential error models, whereas volume of distribution inter-individual variation was omitted after initial iterations yielded negligible values, thereby improving computational convergence and parameter stability. Body weight was incorporated into clearance and distribution volume through allometric scaling using fixed exponents of 0.75 and 1.0, respectively, benchmarked against a 70-kilogram reference individual.
Covariate screening evaluated patient sex alongside continuous clinical variables including age, alanine aminotransferase, blood urea nitrogen, and estimated glomerular filtration rate standardized to cohort medians. Continuous relationships were tested through exponential equations, whereas binary effects were assessed using linear formulations across stepwise forward inclusion and backward elimination routines. The base structural model registered an objective function value of 632.217, with preliminary clearance and volume of distribution calculated at 7.52 liters per hour and 127 liters. Stepwise forward testing revealed that introducing estimated glomerular filtration rate as an exponential covariate produced the most substantial reduction in the objective function value. Subsequent backward elimination confirmed that removing renal filtration caused the objective function value to surge by 52.62, verifying its statistical significance and biological plausibility as a necessary model component.
In the final validated model, clearance was formulated as a function of population typical clearance scaled by weight to the 0.75 power and glomerular filtration rate relative to a cohort median of 173.42 milliliters per minute per 1.73 square meters raised to an estimated power. The typical population values estimated by the final model reached 8.22 liters per hour for clearance and 113 liters for volume of distribution. For a standard 70-kilogram baseline, the resulting clearance normalized to approximately 0.117 liters per hour per kilogram, aligning closely with international pediatric literature benchmarks that span 0.071 to 0.125 liters per hour per kilogram. To ensure computational robustness despite sample size constraints, the proportional residual error was fixed at 0.3 following sensitivity evaluations showing minimal distortion in core parameter estimates.
Model evaluation demonstrated structural consistency and numerical stability across multiple internal validation techniques. Diagnostic goodness-of-fit assessments indicated that observed concentrations matched both population-predicted and individual-predicted values evenly along the line of identity, while conditional weighted residuals remained distributed within two units of zero without apparent trends over time. Non-parametric bootstrap resampling successfully executed 997 out of 1000 iterations, yielding a 99.7 percent robustness rate with parameter medians and 95 percent confidence intervals closely matching NONMEM estimates. Visual predictive checks based on 200 simulated datasets confirmed that observed drug concentrations aligned with simulated fifth and fiftieth percentiles across the dosing window, although wider uncertainty and departures appeared at the 95th percentile due to sparse sampling in upper concentration ranges.
Using the final population parameters, the investigators performed Monte Carlo simulations across 1000 virtual pediatric patients with a median cohort weight of 24 kilograms to evaluate steady-state trough levels under standard guideline recommendations. Simulated scenarios contrasted a regimen of 20 milligrams per kilogram every 8 hours against 15 milligrams per kilogram every 6 hours across normal renal function, mild renal insufficiency, and moderate renal insufficiency. Under both regimens, virtual patients with normal renal function maintained steady-state troughs within the intended 5 to 15 milligrams per liter window. However, patients with reduced renal filtration regularly exceeded the 15 milligrams per liter threshold, with trough concentrations in mild and moderate insufficiency cohorts increasing by 0.8-fold and 1.78-fold under 8-hour dosing, and by 0.83-fold and 1.67-fold under 6-hour dosing.
The authors highlighted several clinical and methodological limitations inherent to the study design. The investigation was retrospective and conducted at a single center without external validation from independent pediatric cohorts, limiting direct extrapolation to broader healthcare environments. Sparse sampling dominated by trough measurements restricted the feasibility of resolving complex multi-compartment kinetics or estimating high-exposure peaks, prompting clinicians to exercise caution when interpreting model predictions exceeding 15 milligrams per liter and discouraging its application above 20 milligrams per liter. Furthermore, fixing residual variability to preserve estimation stability reduced the model’s capacity to describe unexplained data variance. The authors noted that integrating the model into Bayesian therapeutic drug monitoring platforms alongside pharmacodynamic outcome studies will be required before implementing personalized dosing protocols in clinical practice.
The study team selected a one-compartment structural model because the available data consisted primarily of sparse, steady-state trough concentrations. Although a two-compartment model often better characterizes vancomycin pharmacokinetics, the limited sampling made it difficult to reliably estimate volume of distribution and intercompartmental parameters. The authors noted that fixing the volume of distribution was a pragmatic choice to enhance model stability, as their pediatric cohort had unique physiological characteristics that might differ from published literature values. They also fixed the residual error at 0.3 after sensitivity analyses showed that varying this value had minimal impact on the core clearance estimates, a decision made to prevent model convergence issues associated with the small sample size.
Monte Carlo simulations revealed that standard guideline-based dosing regimens often resulted in trough concentrations exceeding the therapeutic window for children with reduced kidney function. For patients with moderate renal insufficiency, the authors suggested specific dose adjustments to mitigate the risk of adverse events. For instance, if the initial regimen was 20 milligrams per kilogram every 8 hours, they proposed reducing it to 15 milligrams per kilogram every 12 hours. Similarly, for a 15 milligrams per kilogram every 6 hours regimen, a reduction to 10 milligrams per kilogram every 8 hours was recommended. These suggestions are based on the model’s predictions of drug accumulation and are intended to help clinicians tailor therapy to individual patient needs.
The model’s predictive accuracy was highest for typical and lower-tail exposure scenarios, but uncertainty increased significantly for high concentrations. The authors cautioned that the model should not be used to predict concentrations above 20 milligrams per liter due to the scarcity of supporting data in that range. This limitation is attributed to the sparse sampling and the fixed residual error, which collectively reduce the model’s precision in estimating extreme values. Clinicians are advised to exercise caution when relying on predicted concentrations that exceed 15 milligrams per liter, as these values may be disproportionately influenced by the limited high-concentration samples available in the dataset.
Future work will focus on developing a Bayesian dose optimization tool that integrates patient-specific covariates and real-time drug concentration data. This tool aims to achieve more accurate individualized dose adjustments by combining the population model with therapeutic drug monitoring. The authors also plan to conduct further research with larger sample sizes to investigate the probability of target attainment for different dosing regimens across various renal function groups. Additionally, they intend to explore pharmacodynamic linkages by combining pharmacokinetic data with bacteriological examinations and clinical outcomes, which could provide a stronger basis for optimizing vancomycin administration in pediatric care.
Subject of Research: Pediatric population pharmacokinetic modeling of vancomycin
Article Title: Establishment of a Vancomycin Population Pharmacokinetic Model for Pediatric Patients Based on the Non-Linear Mixed-Effects Model
Article References: Yu, B., Mei, K., Zhan, D., Tang, Q., Cai, H., & Zhang, R. (2025). Establishment of a Vancomycin Population Pharmacokinetic Model for Pediatric Patients Based on the Non-Linear Mixed-Effects Model. Drugs in R&D, 25(4), 309-320. https://doi.org/10.1007/s40268-025-00523-8
Image Credits: AI Generated
DOI: 10.1007/s40268-025-00523-8
Keywords: Vancomycin, Pediatric Pharmacokinetics, Population Pharmacokinetics, NONMEM, Therapeutic Drug Monitoring, Glomerular Filtration Rate, Dosing Optimization, Nephrotoxicity, Establishment, Population, Pharmacokinetic, Model
News Source: Harold Sullivan. (November 13, 2026). Pediatric Pharmacokinetic Model Pinpoints Weight and Renal Function in Vancomycin Dosing. Scienmag.
