Population Pharmacokinetic Modeling and Target Attainment of Cefepime in Critically Ill Pediatric Patients

Abstract

Background

Cefepime dosing guidelines are well defined in healthy populations but have not been studied extensively in critically ill children who often have significant alterations in antibiotic pharmacokinetics (PK) and pharmacodynamics (PD). Cefepime dosing optimization is important for patient outcomes.

Method

We prospectively enrolled critically ill children receiving cefepime. Plasma samples were collected at opportunistic time points. Cefepime plasma concentrations were measured by liquid-chromatography-tandem mass spectrometry. Population PK (popPK) analysis was conducted with a priori selected covariates. The final covariate model was used to perform probability of target attainment (PTA) analysis.

Results

Data from 84 participants were analyzed. A 2-compartment with proportional error model was selected as the base model. Inclusion of weight and creatinine clearance (CrCl) improved model fit. Estimated breakpoints—the highest minimum inhibitory concentration with ≥90% pharmacokinetic target attainment—were 2-fold higher for 3-hour compared with 30-minute infusions for q8h dosing regimens. For q12h regimens, 3-hour infusions also yielded 2-fold higher breakpoints for the lowest target. Continuous compared with 3-hour infusions resulted in 4- to 16-fold higher breakpoints at the same daily dose at the higher targets. Increased CrCl led to reduced PTA.

Conclusions

Given the variability observed in critically ill children, clinicians should consider incorporating therapeutic drug monitoring and popPK models to determine optimal dosing in this population. The use of extended infusions and more frequent or continuous dosing is beneficial for PK/PD target achievement, particularly for Pseudomonas aeruginosa and antimicrobial-resistant pathogens. Use of these strategies should be strongly considered in the pediatric critically ill population.

This study assessed the population pharmacokinetics of cefepime in critically ill pediatric patients. Weight, creatinine clearance, and infusion time were found to be key drivers of cefepime pharmacokinetics. Moreover, we found that standard cefepime dosing may be insufficient to treat drug-resistant infections adequately.

While dosing of antibiotics has been well studied in healthy populations, few studies have evaluated antibiotic dosing in critically ill pediatric patients. Compared with adults, antibiotic PK is particularly variable in pediatric patients due to age-related alterations in body composition, weight, physiology, and organ maturation [1–4]. Patients who are critically ill may have significant alterations in antibiotic pharmacokinetics (PK) and pharmacodynamics (PD) due to many factors, including hypoalbuminemia, organ dysfunction, augmented renal clearance (ARC) and fluid shifts secondary to fluid resuscitation or systemic inflammatory response syndrome [5–7]. In particular, ARC is a common yet under-recognized phenomenon in critically ill children that can lead to enhanced drug elimination and subtherapeutic antibiotic concentrations due to supraphysiologic renal function [8]. Therapies commonly used in critical care settings, such as extracorporeal membrane oxygenation (ECMO, a modified form of cardiopulmonary bypass) or continuous renal replacement therapy (CRRT), may also impact drug PK and PD through a variety of mechanisms, including increased volume of distribution depletion of plasma proteins, changes in clearance, binding of large molecules to in-circuit filters, and the physiologic effects of the underlying cause of illness, such as sepsis [9, 10]. Critically ill patients are also likely to have more comorbidities compared with the healthy population.

Since critically ill patients are likely to receive multiple courses of antimicrobials, optimization of drug dosing and delivery is critically important for individual patient outcomes and for minimization of adverse drug effects and antibiotic resistance [11–13]. Unfortunately, published studies of antibiotic PK/PD in hospitalized populations are scarce and have focused largely on antibiotics such as vancomycin and gentamicin for which assays for therapeutic drug monitoring (TDM) are widely available [14, 15]. Therapeutic drug monitoring for beta-lactam antibiotics has been challenging historically, due to the lack of validated, commercially available assays. Therefore, only a few studies of cefepime PK/PD have been performed in critically ill pediatric patients and are limited by small sample size [16–19]. Data defining the impact of ECMO or CRRT on cefepime PK in children lags adults [20–25].

The central aim of this study was to develop a population PK (popPK) model to define the PK of cefepime in critically ill children and to identify effective dosing strategies by investigating the effects of dose, dosing interval, infusion time, and other covariates on cefepime PK as well as the probability of attaining PK/PD exposure targets with standard dosing regimens as compared with extended or continuous infusion (CI) regimens.

METHODS

Study Design

A prospective popPK study of beta-lactam antibiotics with opportunistic samples in critically ill children was conducted at Monroe Carell Jr. Children's Hospital at Vanderbilt University Medical Center (VUMC) in Nashville, Tennessee. The VUMC Human Research Protections Program approved the study protocol and amendments. Written informed consent was obtained prior to starting any study related procedures.

Participants

Following informed consent, participants who met eligibility criteria were enrolled between October 2020 and November 2022 at Monroe Carell Jr. Children's Hospital at VUMC. Participants were eligible if they were older than 1 month of age; admitted to the Neonatal Intensive Care Unit, Pediatric Critical Care Unit, or Pediatric Cardiac Intensive Care Unit; and were receiving cefepime. Participants were excluded if they were pregnant; receiving hemodialysis; receiving probenecid; if death was considered imminent (expected within 72 hours); if they were thought by the investigator to be at increased risk by participating; or if they had a comorbidity or other condition that would confound the results of the study.

Drug Administration and Sample Collection

All participants received cefepime per standard of care, as ordered by their treating health care team. In addition to demographic, anthropomorphic, and medical history data, the study team collected information at the time of each dose, including antibiotic of interest, dose, and start and end times of all infusions. Approximately 3–5 mL of blood (calculated based on weight for children such that it would not exceed the lesser of 50 mL or 3 mL/kg in an 8-week period) was collected at each time point. Samples were obtained at times convenient for the health care team, although certain time points were prioritized, such as immediately prior to a dose, at the time of infusion completion, and 1–2 hours after infusion completion. Blood samples were obtained preferentially from a drug-free line, though sampling from the drug infusion line was permitted if a saline flush was administered immediately prior to sample acquisition, and a waste sample was drawn and discarded prior to obtaining the PK sample [26]. Plasma samples were stored at −80 °C until analysis.

Quantification of Plasma Drug Concentrations

A full description of the quantification of plasma drug concentration of cefepime is available in the Supplementary material text and Supplementary Table 1. Briefly, the total plasma concentration of cefepime was measured by a validated liquid-chromatography-tandem mass spectrometry method at the Vanderbilt Mass Spectrometry Core.

Data Processing

Pediatric patients' data were included if their cefepime dosing information occurred within 1 week before enrollment or up to 2 weeks after, yielding a total of 84 patients. Four outlier cefepime blood concentrations were excluded (1.8%) due to suspected line contamination at the time they were obtained.

Population Pharmacokinetic Analysis

We performed popPK analysis using Monolix 2021R^® with the stochastic approximation of expectation-maximization method. First, we selected the base model from 1 and 2 compartment models while assuming log-normal distributions for the random effects on the PK parameters. For each compartment model, we explored models with proportional, additive, and combined error. We selected the best combination of compartment and error models based on the objective function value. After selecting the base model, we performed covariate modeling, starting from modeling weight on PK parameters, which was allometrically scaled to a factor of 70 kg. We used this model as the basis for further covariate modeling to determine what covariates to include in the final model.

The covariate modeling was performed with a priori selected covariates: receiving ECMO, receiving CRRT, sex, race, ethnicity, gestational age, and creatinine clearance. Total protein, alanine aminotransferase (ALT), aspartate aminotransferase (AST), and albumin were not used due to a high proportion of missing data. A sensitivity analysis was conducted on the cohort with complete data for total protein, ALT, AST, and albumin data to ensure reproducibility of results. Continuous covariates, such as gestational age and creatinine clearance were standardized to the cohort's median values. Gestational age was tested as a continuous variable as well as with the age maturation model using the Hill equation. Race, ethnicity, and gestational age each contained missingness for 1 patient, which was imputed with single imputation using the mode for categorical variables and full term (37 weeks) for gestational age.

After selecting a final model, we assessed the model with goodness-of-fit plots, residual plots, prediction-corrected visual predictive check (pcVPC), shrinkage estimates, and normalized prediction distribution errors (NPDEs). The goodness-of-fit plots used population and individual predictions for cefepime blood concentration, and the residual plots used conditional weighted residuals. The pcVPC was employed to understand the uncertainty in model predictions over time. These plots were generated with R programing language (Version 4.4.0, R Core Team, http://www.r-project.org), except for the NPDE graphs that were generated in Monolix.

Probability of Target Attainment Analysis

The final covariate model was used to perform probability of target attainment (PTA) analysis for virtual subjects using Monte Carlo simulation in Simulx^®. First, the covariates were simulated from a log-normal distribution with mean and standard deviation of the study data using R programing language. With simulated covariates, cefepime concentrations were simulated from the final covariate model, which were converted to unbound (free) concentrations using an estimated protein binding of 20% for cefepime [27]. Each subject's dosage was 50 mg/kg with the maximum dose of 2000 mg (ie, if the calculated dose was more than 2000 mg, it was replaced with 2000 mg). With this same dose, impact of infusion duration and dosing frequency on PTA was investigated, yielding 6 dosing regimens: 30 minute standard infusion (SI) given every 8 hours (q8h) or every 12 hours (q12h), 3 hour extended infusion (EI) given q8h or q12h, and 24 hour CI at 100 mg/kg/day and 150 mg/kg/day. Six thousand virtual subjects were simulated for each dosing regimen, with 3 weight categories (≤9 kg, 9–25 kg, and >25 kg) and 2 creatinine clearance categories (≤90 mL/min and >90 mL/min).

The PTA was calculated for each dosing regimen over a grid of possible minimum inhibitory concentration (MIC) values based on 3 targets: 100% fT_>MIC, 65% fT_>MIC, and 100% fT_>4xMIC [28–31]. fT_>MIC indicates the percentage of the dosing interval where unbound plasma drug concentrations exceed the MIC of the causative pathogen. In addition to checking these 3 targets for the dosing regimens, we also considered the target attainment for covariate groups. PK/PD breakpoints, defined as the highest MIC at which 90% of the population is expected to achieve the predetermined PK/PD targets for each dosing regimen, were also calculated.

RESULTS

Study Population and Sample Characteristics

The dataset used for modeling contained 84 patients with 223 cefepime blood concentrations. Table 1 describes the characteristics of the study population and PK data. Age had a strong right skew, with a mean of 6.57 years and a median of 2.61 years, showing that we had more young patients in our dataset. The cohort was predominantly White (81%) and non-Hispanic or Latino (88.1%), with a balanced distribution of sex (49% female). Of the 84 patients, 10 and 9 received ECMO (11.9%) and CRRT (10.7%), respectively. The median number of drug concentrations measured per individual was 2, ranging from 1 to 10, while the median number of dosing events per individual was 11, ranging from 2 to 41. For the sampling times, 16.2% of drug concentrations were considered trough values, defined as measurements within 2 hours pre-dose; 34.5% were peak values, defined as measurements up to 2 hours postdose; and 49.3% were random sampling (Supplementary Table 2).

Table 1.

Characteristics of the Study Population and Pharmacokinetic Data

	Overall (N = 84)
Age (y)
Mean (SD)	6.6 (7.2)
Median [Min, Max]	2.6 [0.1, 26.8]
Gender
Female	41 (48.8%)
Male	43 (51.2%)
Race
Black or African American	11 (13.1%)
Native Hawaiian or other Pacific Islander	1 (1.2%)
Other race	3 (3.6%)
White	68 (81.0%)
Missing	1 (1.2%)
Ethnicity
Hispanic or Latino	9 (10.7%)
Not Hispanic or Latino	74 (88.1%)
Missing	1 (1.2%)
Weight (kg)
Mean (SD)	24.7 (25.5)
Median [Min, Max]	13.7 [3.5, 143.0]
Height (cm)
Mean (SD)	100.0 (41.6)
Median [Min, Max]	87.0 [49.0, 184.0]
Creatinine Clearance (mL/min)
Mean (SD)	105.0 (59.5)
Median [Min, Max]	90.7 [10.8, 275.0]
On ECMO
No	74 (88.1%)
Yes	10 (11.9%)
On CRRT
No	75 (89.3%)
Yes	9 (10.7%)
AST (unit/L)
Mean (SD)	72.4 (174.0)
Median [Min, Max]	39.0 [13.0, 1450.0]
Missing	15 (17.9%)
ALT (unit/L)
Mean (SD)	53.1 (97.1)
Median [Min, Max]	26.0 [6.0, 745.0]
Missing	15 (17.9%)
Albumin (g/dL)
Mean (SD)	3.0 (0.7)
Median [Min, Max]	2.9 [1.6, 4.8]
Missing	8 (9.5%)
Total Protein (g/dL)
Mean (SD)	5.5 (1.3)
Median [Min, Max]	5.3 [3.2, 9.0]
Missing	15 (17.0%)
Gestational age (d)
Mean (SD)	263 (23)
Median [Min, Max]	274 [171, 280]
Missing	1 (1.2%)
Number of drug levels per patient
Mean (SD)	3 (2)
Median [Min, Max]	2 [1, 10]
Number of dosing events per patient
Mean (SD)	14 (9)
Median [Min, Max]	11 [2, 41]
Drug level (mcg/mL)
Mean (SD)	66.1 (44.8)
Median [Min, Max]	60.0 [3.7, 201.0]
Dose (mg/kg)
Mean (SD)	44.4 (9.00)
Median [Min, Max]	47. [11.7, 58.4]

Abbreviations: AST, aspartate aminotransferase; ALT, alanine aminotransferase.

Population Pharmacokinetic Analysis

A 2 compartment with proportional error model was selected as the base model. The random effects on all PK parameters except Q were included, since the random effect of Q could not be estimated precisely, and its inclusion did not improve the model fit. Supplementary Table 3 shows the parameter estimates for the base model, the base model with weight, as well as the final covariate model. Inclusion of weight improved the model fit significantly by reducing the objective function value by 117. Adding creatinine clearance to the model further improved the model fit significantly (objective function value reduction of 64.26).

We observed that receipt of ECMO, sex, and race did not improve the model fit. The use of the age maturation function with gestational age also did not improve the model fit. Receipt of CRRT, ethnicity, and gestational age did individually improve model fit, but their inclusion in the model with weight and creatinine clearance did not provide evidence of being better than the model with only weight and creatinine clearance. The candidate models and their likelihood functions are presented in Supplementary Table 4. The sensitivity analysis of only patients with complete data found the same final model (Supplementary Table 5). The structure of the final covariate model is as follows:

$$\mathrm{C}{\mathrm{L}}_i={\theta }_1\times (\mathrm{w}{\mathrm{t}}_i/70{)}^{0.74}\times (\mathrm{CrC}{\mathrm{l}}_i/90.7{)}^{{\theta }_2}\times \exp ({\eta }_i^{\mathrm{CL}}),$$

$$\mathrm{V}{1}_i={\theta }_3\times (\mathrm{w}{\mathrm{t}}_i/70{)}^1\times \exp ({\eta }_i^{\mathrm{V}1}),$$

$$Q_i={\theta }_4\times (\mathrm{w}{\mathrm{t}}_i/70{)}^{0.75},$$

$$\mathrm{V}{2}_i={\theta }_5\times (\mathrm{w}{\mathrm{t}}_i/70{)}^1\times \exp ({\eta }_i^{\mathrm{V}2}),$$

where wt_i is body weight in kilogram (kg) and CrCl_i is creatinine clearance in mL/min for individual i. CL_i, V1_i, Q_i, and V2_i are the individual specific PK parameter estimates for individual i. Each θ represents a model parameter with ${\theta }_1$, ${\theta }_3$, ${\theta }_4$, and ${\theta }_5$ being population PK parameters of CL (L/hour), V1 (L), Q (L/hour), and V2 (L), respectively, and ${\theta }_2$ being the model parameter for CrCl. The ${\eta }_i^{\mathrm{CL}}$, ${\eta }_i^{\mathrm{V}1}$, and ${\eta }_i^{\mathrm{V}2}$ represent random effects that explain between-individual variability, following a log-normal distribution with means of zero and a covariance matrix that has ${\omega }_{\mathrm{CL}}^2,{\omega }_{\mathrm{V}1}^2,\mathrm{and}{\omega }_{\mathrm{V}2}^2$ along the diagonal.

Figure 1 contains the model assessment results. Figure 1A shows observed versus population (left) and individual (right) predictions, while Figure 1B presents the conditional weighted residuals as a function of time, concentration, weight, and creatinine clearance. Figure 1C shows the VPC plot. Population and individual predicted values in Figure 1A do not stray far from the diagonal line of equality, suggesting a good fit, though there is more variation with larger observed values. The weighted residuals in Figure 1B are approximately symmetric across the horizontal line at 0, indicating a lack of evidence for violation of model assumptions. The VPC in Figure 1C illustrates that the predicted values from the final model align well with the observed values, indicating the model's predictions are reasonable. The NPDEs generally follow a normal distribution (Supplementary Figure 1). Finally, the shrinkage estimate for CL (16.2%) was low, indicating well informed individual-level clearance estimation. In contrast, shrinkage was high for V1 (63.5%) and moderate for V2 (37.0%), suggesting limited information in the data to reliably estimate individual volume parameters (Supplementary Table 6).

Figure 1.

Model diagnostics for final pharmacokinetic model. A, Observed vs predicted concentrations at population and individual levels (left and right, respectively), with identity line (dashed line). B, Individual weighted residuals as a function of time (top left), observed concentrations (top right), weight (bottom left), and creatinine clearance (bottom right), with dashed line at zero and fitted smoother in blue. C, Prediction-corrected visual predictive check with solid lines for empirical percentiles of observed concentration data and shaded areas for the predicted concentration percentiles from simulation. The percentiles used in this visual correspond to the 10th, 50th, and 90th quantiles.

Probability of Target Attainment Analysis

The results from the PTA analysis are shown in Figure 2. We observed an overall ordering of PTA by dosing regimen across all 3 targets, with continuous infusions having the highest PTA, followed by q8h EI, and q12h SI having the lowest PTA. Dosing every 8 hours had increased PTA compared with dosing every 12 hours, and EI also resulted in higher PTA than SI. Figure 3 shows the PTA results by covariate group for the 65% fT_>MIC target. PTA increases with weight and decreases with higher creatinine clearance. Breakpoint values calculated for each dosing regimen are shown in Table 2. Across all targets for the q8h dosing regimens, EI resulted in 2-fold higher breakpoints as compared with SI. For the q12h dosing regimen, EI also resulted in 2-fold higher breakpoints for the 65% fT_>MIC target. For the higher targets of 100% fT_>MIC and 100% fT_>4xMIC, CI resulted in 4- to 16-fold higher breakpoints compared with EI at the same total daily dose of cefepime. The epidemiological cutoff (ECOFF) for cefepime and Pseudomonas aeruginosa is 8 mg/L; therefore, up to this MIC, isolates are considered part of the wild-type distribution [32]. The q8h EI achieved the 65% fT_>MIC target and 2× the ECOFF (16 mg/L) while the SI only reached 1× the ECOFF (8 mg/L) (Table 2). Both CI regimens achieved the 65% fT_>MIC and 100% fT_>MIC targets and 2× the ECOFF. However, no modeled regimen of cefepime achieved the 100% fT_>4xMIC target for the ECOFF.

Figure 2.

Comparison of cefepime probability of target attainment between dosing regimens at each of 3 targets: 100% fT_>MIC (top panel), 65% fT_>MIC (middle panel), and 100% fT_>4xMIC (bottom panel). fT_>MIC indicates the percentage of the dosing interval where unbound plasma drug concentrations exceed the MIC of the causative pathogen. Abbreviations: MIC, minimum inhibitory concentration; q8h, every 8 h; q12h, every 12 h; qd, every 24 h; mg, milligrams; kg, kilograms; L, liters.

Figure 3.

Comparison of cefepime probability of target attainment between dosing regimens for each covariate group at the 65% fT_>MIC target. fT_>MIC indicates the percentage of the dosing interval where unbound plasma drug concentrations exceed the MIC of the causative pathogen. Abbreviations: CrCl, creatinine clearance; MIC, minimum inhibitory concentration; q8h, every 8 h; q12h, every 12 h; qd, every 24 h; wgt, weight in kilograms; mg, milligrams; kg, kilograms; L, liters.

Table 2.

Cefepime Breakpoints (mg/L) for Different Dosing Regimens, Based on Monte Carlo Simulations

Dosing regimen	Cefepime Breakpoints (mg/L) (ie, The Highest MIC At Which ≥90% Of The Participants Achieve Targets)
	65% fT>MIC	100% fT>MIC	100% fT>4xMIC
qd 24-h infusion 150 mg/kg	16	16	4
qd 24-h infusion 100 mg/kg	16	16	4
q12h 3-h infusion	4	1	0.25
q12h 30-min infusion	2	1	0.25
q8h 3-h infusion	16	4	1
q8h 30-min infusion	8	2	0.5

Breakpoints are defined by the highest MIC with ≥90% pharmacokinetic target attainment. fT_>MIC indicates the percent of the dosing interval where unbound plasma drug concentrations exceed the MIC of the causative pathogen.

Abbreviations: MIC, minimum inhibitory concentration; qd, every 24 h; q8h, every 8 h; q12h, every 12 h.

DISCUSSION

In this study, plasma concentrations for cefepime were measured in critically ill pediatric patients and were described by a 2-compartment PK model. Weight had a large impact on cefepime PK, which likely reflects alterations in physiology in the smallest, and often youngest and least developmentally mature, patients [33]. Creatinine clearance also had a large impact on cefepime PK, consistent with the predominantly renal route of elimination of this antibiotic [27]. Other covariates did not substantially affect the PK of cefepime.

Most importantly, our study found that standard dosing of cefepime (50 mg/kg/dose every 8–12 hours, maximum 2 g/dose) may fail to achieve sufficient levels to treat serious infections, particularly in the case of resistant organisms such as P. aeruginosa. Clinically, alterations in cefepime dosing regimen, including frequency of administration and duration of infusion, substantially affected PTA across a broad range of MIC values. Extended infusion times improved the probability of target attainment across all dosing intervals, though did not reach the more aggressive 100% fT_>MIC or 100% fT_>4xMIC targets for the ECOFF for P. aeruginosa. Continuous infusions also improved PTA, even at the more aggressive targets, but still did not reach the ECOFF at the 100% fT_>4xMIC target. Prolonged (extended or continuous) infusion dosing regimens, when feasible, present an attractive option to increase PTA without increasing the total daily dose of cefepime. Currently, there are no established guidelines for use of prolonged infusion regimens of beta-lactams in critically ill pediatric patients due to limited evidence. However, recent consensus guidelines for critically ill adults recommend prolonged rather than SI regimens of beta-lactams in severely ill adult patients, especially those with gram-negative infections, to reduce mortality and increase clinical cure [31]. Given that prolonged infusions of beta-lactams are typically well tolerated and high-quality evidence of the efficacy of prolonged infusions in pediatric patients may be slow to materialize, our results suggest that extended infusions with more frequent dosing or continuous infusions should also be used in critically ill pediatric patients, particularly those with resistant pathogens. Clinicians should incorporate popPK models into their local antibiograms to determine optimal dosing in critically ill pediatric patients, with particular emphasis on patients with altered renal function and low weight since these covariates significantly affect cefepime PK. Renal function may be widely variable in pediatric critically ill patients, including impaired renal function as well as augmented renal clearance (ARC), a phenomenon of supraphysiologic renal function which is increasingly recognized in the pediatric critically ill population [8]. Though we did not specifically evaluate ARC, higher renal clearance was an indicator of decreased PTA in our model.

Prior popPK studies of cefepime in this population are limited. A study of pediatric ICU patients (n = 36) and adult ICU patients used a 2-compartment cefepime popPK model showed that CrCl and weight were significant covariates, similar to our study [16]. Another cefepime popPK study of pediatric ICU patients excluding CRRT recipients (n = 59, 13 of whom received ECMO) similarly showed that weight and renal function (estimated glomerular filtration rate) were significant covariates, though the authors used a 1-compartment model [19]. No popPK models have been developed specifically for pediatric CRRT patients. A small case series (n = 4) showed significant cefepime PK variability in pediatric CRRT patients at more stringent PK/PD targets [21]. A recent study of 7 pediatric CRRT patients similarly showed failure to attain stringent targets [25]. Two additional cefepime PK models in pediatric ECMO patients have been developed. The first study, based on 17 pediatric ECMO patients, was a 2-compartment model that also incorporated weight and serum creatinine with the added covariates of CRRT, blood transfusions, and ECMO-related parameters such as tube coating and oxygenator age [22]. Notably, cefepime measurements demonstrated high variability in individual subjects, with consistently undetectable low levels in 2 of the subjects; subsequent external validation of the model showed poor predictive performance [23]. A new 2-compartment model was developed that included the previous dataset combined with a new dataset (n = 33 pediatric ECMO patients) and included only weight, serum creatinine, oxygenator day, and blood transfusion as covariates [23]. Although we had anticipated that ECMO and CRRT would significantly affect the PK of cefepime in our study, these participants only represented a small subset of our population and are likely to be heterogeneous in terms of other clinical covariates.

Though our study is potentially limited by the sample size, the number of blood samples obtained, and the timing of blood samples for each individual patient, the study's popPK modeling approach helps mitigate these limitations. The combination of compartmental modeling and nonlinear mixed-effects modeling of sparse data drawn from a large and heterogeneous cohort has a unique ability of explaining, identifying, and quantifying the sources of variability that may influence the disposition of the drug. Therefore, our study is likely to provide generalizable and actionable data for these high-risk critically ill patients. While our study assessed creatinine clearance as marker of renal function, this is not a perfect measure of ARC and may be less reliable in critically ill patients with rapidly changing renal function.

Given the baseline variability in clinical covariates observed in pediatric critically ill populations in addition to fluctuations of some of these covariates (eg, renal clearance) over the course of an admission, this population may also benefit from TDM. While use of popPK models and local antibiograms are a useful guide to optimize initial dosing of cefepime, TDM would allow for more personalized antibiotic dosing regimens for this patient population. TDM would also allow institutions without access to local antibiograms or model-based tools to improve beta-lactam dosing. International consensus guidelines for adult critically ill patients already recommend that TDM be performed routinely for beta-lactam antibiotics and pediatric patients may also benefit from this service [34].

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.