Vancomycin Population Pharmacokinetics and Exposure Variability in Children with Multiple Organ Dysfunction Syndrome

Abstract

In children with multiple organ dysfunction syndrome (MODS), vancomycin dosing is frequently guided by population-based approaches followed by therapeutic drug monitoring that requires considerable blood volumes. We sought to develop a population pharmacokinetic (PK) model and describe vancomycin exposures using blood-sparing, whole-blood volumetric absorptive microsampling (VAMS). In this multicentre prospective observational study, up to 15 PK samples were collected per subject over 3 days. Individual whole-blood areas under the concentration-time curve during 0–24 hours (AUC_0–24h) and 24–48 hours (AUC_24–48h) after treatment initiation were estimated. Simulations explored exposures across weight and renal function strata. Sixty-five subjects were included in the final model cohort (median age, 10 years [range, 1 month–17 years]; median weight, 30 kg [3–214 kg]). A two-compartment, allometrically scaled model incorporating renal function best described vancomycin PK. The median dose preceding the first PK sample was 380 mg (15 mg/kg), with most subjects receiving vancomycin every 6 hours. In the fully paired analysis set (n=46), median whole-blood AUC_0–24h and AUC_24–48h were 475 mg·h/L (range, 232–1,749) and 594 mg·h/L (range, 188–2,157), respectively. Simulations demonstrated substantial exposure variability across weight and renal function strata. These findings demonstrate substantial interindividual exposure variability despite population-based dosing. The model provides a foundation for prospective evaluation of individualized exposure estimation using sparse VAMS sampling. Whole-blood VAMS AUC therapeutic targets remain unvalidated, and direct interpretation against established plasma-based targets is an important future effort.

Graphical Abstract

Introduction

Multiple organ dysfunction syndrome (MODS) is a frequent and serious complication of critical illness in children, affecting up to 57% of children in intensive care units [1, 2]. MODS is a potentially reversible condition in which 2 or more organ systems exhibit concurrent dysfunction [1, 3–6]. Among critically ill children, approximately 95% of primary MODS presentations involve a severe, dysregulated systemic inflammatory response [1, 2]. Underlying causes of MODS are varied and include acute illnesses such as sepsis, acute respiratory distress syndrome, and trauma, and MODS often occurs in the context of chronic co-morbidities such as cancer or congenital heart disease [7]. Infection, whether as an inciting factor or as a secondary, nosocomial complication, is also a common contributor, significantly exacerbating morbidity and mortality in this vulnerable population. When MODS is complicated by infection, mortality rates can increase to as high as 67% [8]. A key aspect of optimizing infection outcomes in pediatric MODS is timely, individualized antimicrobial therapy.

Among antimicrobials used in the setting of pediatric MODS, vancomycin is a cornerstone of empiric therapy, due to its activity against methicillin-resistant Staphylococcus aureus (MRSA) [9]. Pediatric vancomycin dosing typically follows weight-based strategies [10, 11], guided by the current Infectious Disease Society of America recommendations to target a 24-hour plasma/serum therapeutic area under the concentration-time curve (AUC) of 400–600 mg·h/L [12]. Notably, exposure:response relationships have demonstrated a clear continuum of toxicity risk as AUC increases [13–15]; however, no such targets exist yet for 24-hour AUC based on whole blood based assays. Achieving therapeutic antibiotic concentrations early in the course of infection is essential for improving survival in pediatric sepsis-induced MODS, yet significant gaps persist in our understanding of how to effectively manage and individualize treatment in this population.

In children with MODS, profound alterations in PK/PD driven by developmental physiology and the extent of organ dysfunction further complicate dosing strategies. Key knowledge gaps remain regarding how antibiotic dosing should be tailored in the setting of MODS. Data from adults suggest that AUC variability is high [16]. Less is known about the AUC variability of children with MODS, and this patient population often provides fewer samples (compared to adults) to guide therapeutic drug monitoring because of blood volume limitations. Approaches such as volumetric absorptive microsampling (VAMS) that require smaller blood volumes for sample collection have become attractive but are less well studied [17].

To address this gap, we conducted a prospective study to develop a population pharmacokinetic (PK) model of vancomycin utilizing VAMS in critically ill pediatric subjects with MODS and describe the variability of AUC achieved with current dosing strategies in this population.

Materials and Methods

Study Design, Oversight, and Enrollment

The Assessment of MODS: PersonaLizing Exposures (AMPLE, R01HD103755, MPI: Downes, Scheetz) study is a multi-center, prospective, observational PK study focused on quantifying antibiotic exposures in the setting of MODS. AMPLE is ancillary and embedded within the PediAtric ReseArch of Drugs, Immunoparalysis and Genetics during MODS (PARADIGM, R01HD095976, PI: Hall) study, which aims to study the immune function and outcomes in children with MODS conducted across the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) network. The AMPLE study is conducted across 15 pediatric intensive care units (PICU) across the United States in California, Colorado, Minnesota, Iowa, Arkansas, Indiana, Michigan, Ohio, Pennsylvania, Maryland, and Delaware.

To be eligible for PARADIGM, children (<18 years of age) must be hospitalized in the PICU with acute MODS onset within the past 3 calendar days (complete eligibility criteria are shown in Table S1). Children enrolled in PARADIGM who received at least 1 of 6 qualifying intravenous (IV) antibiotics were eligible for participation in AMPLE (see Table S1). Enrollment in AMPLE began in June 2022 and is currently ongoing. Data handling and covariate preparation followed the previously reported PARADIGM/AMPLE workflow, with manuscript-specific exceptions described below.[18] This report is an a priori planned population PK model for vancomycin after enrollment of ≥30 subjects, a threshold selected to support pediatric population PK model development and parameter-precision assessment rather than conventional hypothesis-testing power calculations.[19] For the current analysis, AMPLE subjects with complete data available who were administered IV vancomycin, had ≥1 PK sample collected, and did not receive continuous renal replacement therapy (CRRT) or plasmapheresis during the study period were included. Written parent/legal guardian informed consent (and where appropriate, subject assent) was obtained by study staff prior to study participation. Consent for AMPLE was embedded within the PARADIGM consent process. Nationwide Children’s Hospital serves as the central IRB (19–016533) for both studies.

Vancomycin Sampling

Antibiotic selection and dosing was at the clinicians’ discretion. PK samples were collected as soon as possible after MODS onset, once the subject was on ≥1 eligible antibiotic. Those who received IV vancomycin and had ≥1 PK sample obtained around the period of MODS were eligible for inclusion in the current analysis (Figure S1) with PK samples obtained within 3 days of AMPLE eligibility. On each day, blood was sampled up to 5 times for a maximum of 15 samples per subject.

Volumetric Absorptive Microsampling and Spectrometry

Blood samples were obtained via arterial or venous catheter (that was not used for antibiotic administration) or capillary puncture. 20 μL of whole blood was loaded to Mitra^® tip volumetric absorptive microsampling (VAMS) devices (Mitra^® by Trajan Scientific and Medical, Victoria, Australia). Each blood sample was obtained in duplicate (i.e. 40 μL in total was collected per sample). VAMS collection devices were dried at room temperature for 1–3 hours, refrigerated (i.e. 4° C) for a maximum of 72 hours, and then transferred to a storage temperature of −80° C for batch analysis. Our previous stability studies demonstrated that this protocol enables analyte stability for up to 90 days [20]. Samples were transferred to a central laboratory (University of Michigan) on dry ice. Sample analysis was performed on ultra-high performance liquid chromatography-tandem mass spectrometry (Shimadzu, Kyoto, Japan) as previously described [20]. Briefly, separation was achieved on a Phenomenex (CA, USA) Kinetex^™ Polar C18 column (4.6 mm × 50 mm, 2.6 μm) with C18 Polar 3.0-mm ID guard column. Vancomycin was detected with AB Sciex (CA, USA) 4500 triple quadrupole mass spectrometer. This method allowed multiplex of all AMPLE analytes from the 20 μL microsamples and was validated according to U.S. FDA guidelines for bioanalytical method validation [20]. Because internal data demonstrated inaccuracy for hematocrits less than 22%, vancomycin concentrations collected while a subject’s hematocrit was below 22% were censored.

Population Pharmacokinetic Model Development

Parametric population pharmacokinetic (popPK) modeling and simulation were performed using Monolix and Simulx 2024R1 (Lixoft, Antony, France), respectively for all subjects with at least one vancomycin dose and measurable concentration. Concentration records were excluded from the pharmacokinetic analysis when the subject’s most recently recorded hematocrit was <22%. Within each subject, concentration records were excluded beginning at the first hematocrit value <22% and continuing until, but not including, the first subsequent hematocrit value >22%. Dose records were retained during these intervals to preserve the administered dosing history for pharmacokinetic model fitting. Base structural models evaluated and tested 1- and 2-compartment structural models with first-order elimination. Concentrations that were below the lower limit of quantification (i.e. <1 mg/L) were handled with Beal method 4 [21]. Random effects were estimated for clearance (CL), central volume compartment (V1) [1-compartment models] as well as peripheral volume compartment (V2), and intercompartmental transfer (Q) [2-compartment models]. The base model was selected by improvement in the corrected Bayesian Information Criterion (BICc).

Stepwise assessment of covariate effects on fixed effects parameters was then performed. First, compartment and error structures were evaluated. Standard errors were estimated from Fisher Information Matrix using stochastic approaches. Constant, proportional, and combined error models were tested to evaluate residual variability. Then the impact of allometric scaling on model fit was assessed using factors with powers of 0.75 and 1 for CL and volume terms respectively [22]. Lastly, the influence of covariates was tested on model parameters. As appropriate, time-varying covariates that were evaluated included estimated glomerular filtration rate (eGFR), calculated by modified Schwartz and the Chronic Kidney Disease in children Under 25 (CKiD U25) equations [23–26] and laboratory data including serum creatinine and tumor necrosis factor alpha (TNF-α) stimulation response. Immunoparalysis status was defined as tumor necrosis factor alpha (TNF-α) stimulation response <200 pg/mL [27–29]. Time-varying laboratory covariates, including eGFR, were interpolated between measured values and handled as regressors in the Monolix workflow. Additional covariates evaluated included age, weight, height, sex, vasopressor use, ventilator use, fluid status, and daily organ function scores including the Pediatric Logistic Organ Dysfunction-2 (PELOD-2) [30], PROULX [31], and Pediatric Risk of Mortality (PRISM) scores [32]. When height was missing, the CDC 50th percentile height for age and sex was used. Daily organ function scores were applied by calendar day. For time periods preceding MODS onset, organ function scores were assigned as one-half of the MODS day 0 value. When baseline covariate values were unavailable, the first observed value was carried backward to time zero; subsequent missing values were carried forward to the next observed value when applicable. Covariates were retained based on reductions in the Corrected Bayesian Information Criteria (BICc), improved precision of parameter estimates, reduced interindividual variability, and visual improvements in goodness-of-fit plots.

Vancomycin AUC Calculations

Whole-blood vancomycin areas under the concentration-time curve during 0–24 hours (AUC_0–24h) and 24–48 hours (AUC_24–48h) after treatment initiation were estimated for each subject using empirical Bayes estimates (EBEs), administered doses, and modeled concentrations. AUCs were calculated with respect to the start of vancomycin therapy for each individual regardless of the timing of vancomycin administration relative to MODS onset. Because subjects received vancomycin for varying durations and at varying dosing frequencies, AUC summaries were restricted to subjects with sufficient dosing exposure in both calculation windows. Subjects were included in the paired AUC summary only if they received at least two vancomycin doses during the first 24-hour period after treatment initiation and at least two vancomycin doses during the 24- to 48-hour period.

Vancomycin Simulations

The final popPK model was leveraged to generate population simulations for vancomycin AUC_0–24h and AUC_24–48h. Simulations represented 1,000 virtual individuals for each weight/GFR stratum. Stratifications were created for different weights and eGFR. Weight stratification categories were set according to quartiles from the original data set (i.e. 16 kg, 30 kg, and 55 kg). eGFR categories were set to common clinical dosing stratifications of 30, 60, 90, and 120 mL/min. The vancomycin dose applied was set at 15 mg/kg every 6 hours for all dosing weights and GFRs of 60, 90, and 120. For poor renal function status (e.g. eGFR 30 mL/min), the dosing was 15 mg/kg every 12 hours based upon current drug information resources (i.e. Lexicomp) and institutional practices [33]. These regimens were selected as clinically interpretable population-dosing reference scenarios for evaluating exposure variability and were not intended as an exhaustive dose-optimization analysis.

Model Evaluation and Diagnostics

Model adequacy was evaluated using graphical, numerical, and simulation-based diagnostics before interpretation of final exposure simulations. Diagnostic assessment included observed-versus-population and observed-versus-individual predicted concentration plots, individual weighted residuals versus time and individual predictions, residual distribution diagnostics, conditional-mode shrinkage estimates for random effects, and precision of fixed and random effects. Predictive performance was evaluated using a prediction-corrected visual predictive check generated from the final model with 500 simulated replicates, with empirical percentiles compared against model-based 90% prediction intervals. Normalized prediction distribution error diagnostics were also evaluated by comparing empirical NPDE distributions and cumulative distributions with theoretical standard normal distributions.

Results

Study Population

Demographic and clinical summaries follow and are in Table 1. A total of 66 critically ill children met inclusion criteria and contributed 625 vancomycin concentration-observation records before hematocrit-based exclusions. One child had no evaluable vancomycin observations remaining after these exclusions; therefore, 65 subjects contributed to the final model. Among these subjects, thirty-four observation records occurred during hematocrit-defined VAMS unreliability intervals and were excluded, leaving 591 observations for pharmacokinetic model development. Thirty-five of the 591 observations included in the model were below the lower limit of quantification.

Table 1.

Baseline characteristics of the final model cohort

Characteristic	N = 65, Median (IQR)
Age (yrs)	10.03 (4.49 – 13.86)
Weight (kg)	30.00 (16.20 – 59.50)
Height (cm)	127.22 (100.20 – 155.00)
PELOD Scores (pre first PK)	6.00 (5.00 – 9.00)
PROULX Score (pre first PK)	2.00 (2.00 – 3.00)
PRISM Score (pre first PK)	10.00 (3.00 – 16.00)
SCr (mg/dL) (pre first PK)	0.60 (0.35 – 0.80)
Modified Schwartz eGFR (mL/min/1.73 m2) (pre first PK)	95.00 (65.00–131.00)
U25 eGFR (mL/min/1.73 m2) (pre first PK)	92.00 (62.00–118.00)
MODS day on first PK sample	1.00 (1.00 – 2.00)
Vancomycin Dose preceding first sample (mg/kg)	15.00 (12.06 – 17.44)
Female Sex, n (%)	36 (55.4)
Baseline Vasopressors (Yes), n (%)	55 (84.6)
Baseline Mechanical Ventilation (Yes), n (%)	63 (96.9)

Abbreviations

IQR: Interquartile range (i.e. 25^th percentile – 75^th percentile)

PELOD: Pediatric Logistic Organ Dysfunction (Score)

PROULX: Proulx (Score)

PRISM: Pediatric Risk of Mortality (Score)

SCr: Serum creatinine

eGFR: Estimated glomerular filtration rate

The median vancomycin dose that preceded the first sampling event was 15 mg/kg, and the first sampling occurred at a median of MODS day 1 (Figure 1). Dosing frequency was as follows. A total of 30 subjects received vancomycin every 6 hours, 21 subjects every 8 hours, and 10 subjects every 12 hours. Four subjects received only one dose. The median age was 10.0 years (IQR: 4.49 – 13.86), median weight was 30.0 kg (IQR: 16.20 – 59.50), and median height was 127.2 cm (IQR: 100.20 – 155.00). Thirty-six participants (55.4%) were female. Fifty-five (84.6%) subjects were on vasopressors and 63 (96.9%) were receiving mechanical ventilation prior to the first PK sample. The median PELOD-2 [30] and PRISM III [32] scores were 6.0 and 10.0, respectively. Pre-PK median serum creatinine was 0.60 mg/dL (IQR: 0.35 – 0.80). Median pre-PK eGFR was 92.0 mL/min/1.73 m² (IQR: 62.00 – 118.00) by CKiD U25 and 95.0 mL/min/1.73 m² (IQR: 65.00 – 131.00) by modified Schwartz.

Figure 1.

Distributions of (A) MODS day at the first PK sample and (B) vancomycin dose preceding the first PK sample in the final model cohort (n=65).

Population PK Model

The model-building path and covariate assessment are summarized in Table S2. Final model diagnostic outputs are provided in Supplementary Figures S2–S4 and Table 2. A two-compartment model improved BICc relative to the one-compartment model (3527.05 vs. 3628.08). Allometric scaling further improved BICc to 3425.46, and a proportional residual error model improved BICc to 3414.60. Addition of CKiD U25 eGFR on clearance produced the final model and improved BICc to 3379.18 (OFV 3324.2). No other evaluated covariates improved model fit.

Table 2.

Population PK parameter fixed effects estimates and random effects (i.e. between subject variability) for central volume of distribution (V1), clearance (CL), beta coefficients on weight-adjusted (individual weight/28.4 kg) for V1 and CL, peripheral volume of distribution (V2), and intercompartmental bi-directional flow (Q). The final model was developed using 65 subjects after exclusions.

Parameter	Value	C.V. (%)	Stoch. Approx.				Cond. Mode
			Standard Error	Relative Standard Error (%)	P2.5	P97.5	Shrinkage (%)
Fixed Effects
CLi (L/h)	1.996		0.143	7.173	1.736	2.297	7.758
θ_U25 GFR	0.671		0.125	18.569	0.427	0.915	N/A
V1 i (L)	8.469		0.945	11.161	6.818	10.519	45.895
Q i (L/h)	2.369		0.723	30.517	1.350	4.157	26.167
V2 i (L)	13.201		1.395	10.564	10.751	16.210	51.015
Standard Deviation of the Random Effects
ω_CL	0.546	58.915	0.0546	10.008	0.449	0.663
ω_V1	0.493	52.430	0.094	19.068	0.342	0.709
ω_Q	1.422	255.893	0.310	21.799	0.940	2.149
ω_V2	0.443	46.612	0.0881	19.875	0.304	0.648
Error Model Parameters
b	0.241		0.00897	3.725	0.224	0.259

Model Equations:
$C{l}_i=Cl*{\left(\frac{Wt}{28.4}\right)}^{0.75}*{\left(\frac{U25}{96}\right)}^{\theta }GFR*e^{{\eta }_{CL}}$	$Q_i=Q*{\left(\frac{Wt}{29.8}\right)}^{0.75}*e^{\eta Q}$
$V{1}_i=V1*{\left(\frac{Wt}{28.4}\right)}^1*e^{{\eta }_{V1}}$	$V{2}_i=V2*{\left(\frac{Wt}{28.4}\right)}^1*e^{\eta V2}$

The final model was an allometrically scaled two-compartment model with CKiD U25 effects on clearance (Table 2). Final population parameter estimates were CL 1.996 L/h, V1 8.469 L, Q 2.369 L/h, V2 13.201 L, and theta_U25 GFR 0.671. Relative standard errors were 7.2% for CL, 11.2% for V1, 30.5% for Q, 10.6% for V2, and 18.6% for theta_U25 GFR.

Vancomycin AUCs From the Study Population

Because subjects received vancomycin for varying durations and at varying dosing frequencies, model-derived AUC summaries were restricted to subjects with at least two vancomycin doses in each 24-hour calculation window. Forty-six subjects met this fully paired criterion and were included in both the AUC_0–24h and AUC_24–48h summaries. The median whole-blood AUC_0–24h and AUC_24–48h values were 475 mg·h/L (range: 232–1,749) and 594 mg·h/L (range: 188–2,157), respectively (Figure 3).

Figure 3.

Individually calculated 24-hour whole-blood VAMS AUC values from Simulx using empirical Bayes estimates. AUC summaries were restricted to the fully paired subset of subjects who received at least two vancomycin doses in both the 0–24 h and 24–48 h calculation windows (n=46). Panels show (A) AUC_0–24h histogram, (B) AUC_24–48h histogram, (C) boxplot of AUC_0–24h and AUC_24–48h, and (D) paired subject-level AUC changes. AUC_0–24h represents the second 24-hour interval and is not cumulative AUC₀–₄₈h. These whole-blood VAMS AUCs are descriptive; validated therapeutic AUC targets have not been established.

Vancomycin Simulations for Representative Weight Categories and Glomerular Filtration Rate (GFR) Strata

Simulations demonstrated substantial exposure variability across all evaluated weight and renal function strata (Figure 4). Across AUC_0–24h strata, median simulated exposures ranged from 451 to 755 mg·h/L. The highest median AUC_0–24h occurred in the 55 kg, GFR 60 mL/min group, with a median of 755 mg·h/L and a range of 223–2,187 mg·h/L. Across AUC_24–48h strata, median simulated exposures ranged from 595 to 1,149 mg·h/L. The highest median AUC_24–48h also occurred in the 55 kg, GFR 60 mL/min group, with a median of 1,149 mg·h/L and a range of 248–3,517 mg·h/L.

Figure 4.

Simulated vancomycin (A) AUC_0–24h and (B) AUC_24–48h values stratified by weight quartile and glomerular filtration rate (GFR) strata. Each boxplot represents 1,000 simulated individuals. These whole-blood VAMS AUCs are descriptive; validated therapeutic AUC targets have not been established.

Model Evaluation and Diagnostics

Goodness-of-fit diagnostics supported acceptable model performance for the final population pharmacokinetic model. Individual weighted residuals were generally centered around zero without a dominant trend over time or across individual predictions, although several outlying residuals were observed (Figure S2). The prediction-corrected visual predictive check showed that the empirical percentile trends were generally contained within the model-based 90% prediction intervals across time after last dose, supporting the predictive performance of the model; early post-dose percentile estimates included limited outlier flags (Figure S3). Residual distribution diagnostics showed that IWRES were approximately centered and closely followed the theoretical distribution, whereas NPDE were centered near zero but showed modest deviation from the theoretical standard normal distribution (Figure S4). Conditional-mode shrinkage was low for clearance, moderate for intercompartmental clearance, and high for the volume parameters, with shrinkage values of 7.8% for CL, 26.2% for Q, 45.9% for V1, and 51.0% for V2. Parameter precision was acceptable for CL, V1, V2, the U25 CKiD effect, and the residual error term, while Q was estimated with lower precision than the other fixed effects.

Discussion

Despite pediatric subjects with MODS having median VAMS-based AUCs on day 1 and day 2 of 475 and 594 mg·h/L, respectively, daily AUCs ranged from 188–2,157 mg·h/L over 48 hours. This variability supports the need to better control vancomycin exposures in pediatric patients with MODS. When comparing pediatric subjects with MODS to a more general group of children in intensive care units, our range estimations of AUC are wider [34], reflecting the greater impact of end-organ dysfunction in the pediatric population with MODS. Current vancomycin dosing strategies in pediatric patients initially involve weight-based methods [10, 11] that are later guided by therapeutic drug monitoring. VAMS concentrations that relate to efficacy and toxicity will need to be formally defined.

Our popPK model parameter estimates are grossly similar but do differ from those reported in the literature for critically ill children. For example, we found mean parameter values of 0.07 L/kg/h and 0.3 L/kg for CL and V respectively. Others have shown parameter values for CL and V to range from 0.014–0.27 L/kg/h and 0.43–1.46 L/kg, respectively [35, 36]. One potential reason for differences may be the high variability in drug clearance and distribution in children with MODS. Additionally, our model was derived using whole blood VAMS whereas most clinical studies use plasma or serum concentrations. VAMS is a relatively novel approach to obtaining dried blood samples for quantitative analysis of drugs [17, 37, 38]. This technique is promising, especially in pediatric patients where multiple collections of blood are needed to conduct clinical studies and analysis for therapeutic drug monitoring. Traditionally, large volumes of blood are needed to measure vancomycin levels, but VAMS enables the collection of small volumes (e.g. 10, 20, or 30 μL) of blood with high accuracy. Furthermore, while other collection techniques such as dried blood spot also enable smaller volume sizes (e.g. 30 μL per spot), limitations include the impact of varying hematocrit levels on the analysis and performance. Utilizing the VAMS approach partially mitigates against effects of hematocrit on sample analysis and other complications. Direct correlations of VAMS-derived AUCs to the IDSA-recommended plasma AUC targets of 400 – 600 mg·h/L are not possible at this time. While the exact relationship between plasma and VAMS-based concentrations are not clear yet (and is undergoing study), it is clear from the individual PK plots that drug exposure and elimination is well modeled (i.e., that the individual PK is explainable at the subject level and there is not random variation at the VAMS level that prevents modeling because of unexplained variability). Thus, these PK data will serve as an important first step to define VAMS PK until targets can be directly linked to plasma PK.

Whole-blood VAMS concentrations are not directly equivalent to serum or plasma concentrations. In one study, vancomycin VAMS concentrations were approximately 23% to 36% lower than paired plasma concentrations, corresponding to VAMS:plasma ratios of 0.64 to 0.77 [39]. These differences likely reflect limited vancomycin partitioning into erythrocytes, although hematocrit, matrix effects, and analytical recovery may also influence the observed relationship. In a separate laboratory-based paired-sample study, we observed a mean vancomycin VAMS:plasma ratio of 0.62, with ratios ranging from 0.60 to 0.64 across the tested concentrations [40]. However, clinical validation is needed before VAMS concentrations can be routinely converted to plasma-equivalent concentrations or interpreted against established plasma-based therapeutic targets.

Our results are important as pediatric patients with MODS are recognized to be among the most vulnerable, and few studies exist to guide vancomycin dosing in this population. A 2017 study by Doctor, et al highlighted the need for a better understanding of PK and pharmacodynamics (PD) in children with MODS, [41] however the study did not provide specific recommendations regarding antibiotic dosing. Similar to children in our study, adults with MODS display significant changes in volume of distribution (Vd) and CL [42]. These alterations often result in subtherapeutic antibiotic concentrations, particularly in the initial 24 hours of therapy. The authors recommended “front-loading” antibiotic doses to overcome this challenge and adjusting maintenance doses according to changes in clearance associated with evolving organ dysfunction. However, this guidance does not directly address the unique and dynamic physiology of pediatric patients.

The vancomycin simulations demonstrated substantial variability in model-predicted exposure across the evaluated weight and GFR strata (Figure 4). Across strata, median simulated AUC_0–24h values ranged from 451 to 755 mg·h/L, and median AUC_24–48h values ranged from 595 to 1,149 mg·h/L. Variability was evident within each simulated subgroup, with wide exposure ranges despite fixed weight and GFR category assignments. For example, among simulated individuals weighing 30 kg with a GFR of 90 mL/min, a subgroup approximating the median weight and renal function of the study cohort, the median AUC_0–24h was 606 mg·h/L with a range of 170–1,769 mg·h/L, while the median AUC_24–48h was 820 mg·h/L with a range of 182–2,771 mg·h/L. These findings emphasize that substantial exposure heterogeneity persisted even after stratification by body size and renal function.

Our study and analysis are subject to several limitations. First, blood sampling occurred across 3 consecutive days, but the sampling schedule was not strictly aligned with MODS onset. Measurement of vancomycin at different points in MODS progression had the potential to introduce additional variability into our current analysis. Subjects were represented across the continuum of MODS and vancomycin start times. Yet our study benefited from a careful prospective sampling scheme. Renal function estimation in the youngest subjects should also be interpreted cautiously, as the modified Schwartz and CKiD U25 equations have limited validation in infants and only 6 evaluable subjects were younger than 1 year. We evaluated postmenstrual age–based maturation effects, maturation without the eGFR renal function term, and an empirical piecewise clearance term for children younger than 2 years; however, these sensitivity analyses did not meaningfully improve model performance or produced unstable parameter estimates (data not shown). Therefore, a separate maturation term was not retained in the final model. Conditional-mode shrinkage was relatively high for V1 (45.9%) and V2 (51.0%), indicating limited individual-level information about distribution volumes and greater reliance of their EBEs on population estimates. Individual volume estimates and EBE-based covariate or parameter-correlation diagnostics should therefore be interpreted cautiously. Simulation-based diagnostics complement these individual-fit assessments. Although shrinkage for CL was low (7.8%), this does not guarantee that early, non-steady-state AUC estimates are unaffected by uncertainty in distribution parameters. Finally, only VAMS-based PK data are presented at this time; however, kidney outcomes will be described in future efforts.

Our model provides a foundation for prospective evaluation of Bayesian exposure estimation using sparse VAMS sampling. Clinical dosing applications require prospective validation and establishment of whole-blood VAMS exposure–response relationships or a clinically validated translation to plasma-equivalent exposures, including evaluation of kidney outcomes. In conclusion, we developed the first (to our knowledge) vancomycin VAMS-based popPK model in critically ill children with MODS. The model incorporated a 2-compartment model that was allometrically scaled and included CKiD U25 effects on CL. Children with MODS had widely varying vancomycin AUCs based upon actual doses received or simulated initial dosing approaches. This study demonstrates substantial interindividual AUC_0–24h variability despite population-based dosing approaches, suggesting that population strategies alone are insufficient to reliably control individual exposure early in MODS. Our model can serve as a Bayesian prior to optimize individual patient exposures (i.e. deliver personalized, precision dosing to this patient population) with sparse clinical sampling. Future prospective validation is needed, including evaluation of kidney outcomes.

Figure 2.

Observed whole-blood VAMS vancomycin concentrations versus (A, left) population predictions and (B, right) individual predictions for the final population PK model. Concentrations and predictions are expressed in mg/L. These are model-fit diagnostics, not assessments of therapeutic target attainment.

Highlights.

• Volumetric Absorptive MicroSampling (VAMS) is a novel whole-blood collection technique with minimal sampling volumes.

• An allometrically scaled 2-compartment popPK model parameterized with renal function describes individual patients well in children with multiple organ dysfunction syndrome (MODS).

• Simulations demonstrate that population-based starting regimens result in large variability in children with MODS.

• Interindividual vancomycin AUC_0–24h is highly variable despite population-based dosing approaches.

• Individualized approaches are needed for vancomycin dosing in MODS.

Resource Sharing Statement

Data are available through NICHD’s Data and Specimen Hub (DASH) at the following link: https://dash.nichd.nih.gov/study/103755. This page provides study information and instructions for accessing the dataset. Note that at the time of manuscript acceptance, NICHD is/was in the process of migrating DASH to a new platform. A link to the new data repository will be provided when available.