Effect of Cystatin C on Vancomycin Clearance Estimation in Critically Ill Children Using a Population Pharmacokinetic Modeling Approach

Abstract

Background.

Vancomycin is eliminated via glomerular filtration, but current approaches to estimate kidney function in children are unreliable. The authors sought to compare the suitability of cystatin C (CysC)-based glomerular filtration rate equations with the most commonly used creatinine-based equation, bedside Schwartz, to estimate vancomycin clearance (CL).

Methods.

This prospective observational study enrolled critically ill patients (2–18 years) receiving intravenous (IV) vancomycin at the Children’s Hospital of Philadelphia during December 2015-November 2017. Vancomycin levels were collected during clinical care and at 3 times during a single dosing interval. Plasma CysC was measured within 24 hours prior to IV vancomycin (baseline) initiation or immediately following enrollment, as well as along with the third pharmacokinetic (PK) sample. Nonlinear mixed effects modeling was performed using NONMEM software. Covariate selection was used to test model fit with inclusion of estimated glomerular filtration rate (eGFR) on CL using bedside Schwartz versus various published CysC-based equations.

Results.

In total, 83 vancomycin levels were obtained from 20 children. Median age was 12.7 years; 6 patients were female. A one-compartment model best described the data; CL was allometrically scaled to 0.75. During covariate selection, inclusion of eGFR calculated using a CysC-based equation significantly improved model fit (reduction in objective function value [OFV] range: −17.191 to −18.704) than bedside Schwartz (ΔOFV −12.820). Including the full age spectrum equation, an eGFR equation based on both creatinine and CysC, led to the largest OFV reduction (−22.913); female sex was also a significant covariate of CL in the model. Final model pharmacokinetic indices were CL = 0.29 L/hr/kg^0.75 and volume of distribution = 0.48 L/kg.

Conclusions.

CysC-based equations help better estimate vancomycin CL than bedside Schwartz in critically ill children.

Vancomycin is a glycopeptide antibiotic and the drug of choice for empiric treatment of serious Gram-positive infections. Delayed attainment of adequate vancomycin concentrations is associated with treatment failure in adults with methicillin-resistant Staphylococcus aureus infections.^1–4 However, vancomycin has a narrow therapeutic index and exhibits area under the curve (AUC)-dependent efficacy and nephrotoxicity^5–7; an AUC_0–24 ≥800 mg*hr/L carried an adjusted odds ratio of 3.7 for acute kidney injury (AKI) in one large pediatric study.⁵ Critically ill children sustain AKI during vancomycin therapy in 15–25% of courses,^6–8 which independently increases mortality and is associated with development of chronic kidney disease.^9–11 Therefore, a balance exists between maximizing efficacy for potentially life-saving treatment and minimizing toxicity.

Vancomycin is eliminated via glomerular filtration,¹² thus total body clearance (CL) correlates with glomerular filtration rate (GFR). Direct measurement of GFR in the critical care setting is impractical and clinicians utilize surrogate markers, such as serum creatinine, to estimate GFR (eGFR). Unfortunately, creatinine is a suboptimal marker of kidney function in children as values are affected by age, sex, medications, and muscle mass,¹³ and it has a very long serum half-life when renal function is impaired, requiring days to reach steady-state in the case of AKI.¹⁴ Although creatinine-based GFR equations poorly predict vancomycin trough values in children,¹⁵ they are still commonly used for dosing guidance in this setting. Bedside Schwartz equation (Schwartz_bed),¹⁶ which requires serum creatinine and height for calculations, is utilized for GFR estimation in children at our hospital, as well as many others.

Cystatin C (CysC) is an endogenous cysteine protease inhibitor widely expressed by nucleated cells and produced at a constant rate in the body.¹⁷ It is freely filtered across the glomerular membrane, and its increased blood concentrations are almost exclusively due to reduced glomerular filtration. Studies have described its superiority over creatinine for GFR estimation and AKI detection in critically ill children.^18–21 Adult studies have demonstrated improved correlation between CysC and vancomycin CL,^22,23 as well as vancomycin troughs,^24,25 compared with creatinine. However, fewer studies have been performed in pediatric patients,²⁶ and the relationship between CysC and vancomycin CL in critically ill children, specifically, has not been described.

The availability of numerous creatinine- and CysC-based GFR estimating equations makes it challenging for clinicians to select the way to evaluate kidney function in critically ill children. This may lead to over- or under-dosing of vancomycin. We sought to compare eGFR derived from the creatinine-based Schwartz_bed equation,¹⁶ the most common clinically used equation to estimate GFR in children, with eGFR calculated using various published CysC-based equations (Table 1)^27–32 as covariates on vancomycin CL in a population pharmacokinetic (PK) model of vancomycin in critically ill children.

Table 1.

Published equations for estimating GFR evaluated in this study

Equation/author name	Equation for estimating GFR
Creatinine-based equation(s)
Schwartzbed16	0.413*(Ht/Cr)
CysC-based equation(s)
Schwartzcysc29	40.6 * (1.8 / CysC)0.93
CKD-EPI32	If CysC ≤ 0.8 mg/L: 133 *(CysC / 0.8)−0.499 * 0.996age [*0.932 if female] If CysC > 0.8 mg/L: 133 *(CysC / 0.8)−1.328 * 0.996age [*0.932 if female]
Hoek et al.28	−4.32 + (80.35 * CysC)
Grubb et al.31	130*(CysC−1.069) * age−0.117 – 7
Zappitelli et al.30	75.94 / (CysC1.17)
Combined creatinine and cystatin C-based equation(s)
Full age spectrum (FAScomb)27	107.3 / [(Cr/QCr + CysC/QCysC) / 2]a

^aQ_Cr is the mean or median Cr value of the 1-year age-specific distribution of Cr in healthy children. Q_cysc is 0.82 mg/L.

Cr, creatinine; CysC, cystatin C; GFR, glomerular filtration rate; Ht, height

Materials and Methods

STUDY POPULATION

This was a prospective, observational PK study conducted in the Pediatric Intensive Care Unit at the Children’s Hospital of Philadelphia (CHOP) from December 2015 to November 2017. Patients aged 2–18 years receiving intravenous (IV) vancomycin for any indication were eligible for inclusion. Subjects were excluded if they a) had a preceding diagnosis of hypo- or hyperthyroidism, since this can affect CysC level,³³ or b) if they were receiving renal replacement therapy or extracorporeal membrane oxygenation. The CHOP Institutional Review Board approved the study protocol with a waiver of documented assent; verbal assent was obtained, as appropriate. Written informed consent was obtained from the subject’s parent/legal guardian. The study was conducted in accordance with criteria set by the Declaration of Helsinki. Samples collected specifically for the study were paid for by the research team while clinically obtained samples were billed to the subject’s insurance.

DOSING, PK SAMPLING, AND BIOMARKER MEASUREMENT

Vancomycin dosages and infusion rates were determined by the clinical team caring for the subject. Typical initial dosages administered at our institution were 10–15 mg/kg every 6–8 hours, infused over 1–2 hours, and were adjusted to age, weight, and estimated renal function. Vancomycin trough levels (C_min) were routinely obtained <30 minutes prior to the fourth dose in subjects with expected treatment time of >48 hours, and dosages were adjusted to attain a target vancomycin trough concentration of 10–20 mg/L. All vancomycin levels obtained clinically were recorded. Three additional study-directed vancomycin concentrations were obtained during a single dosing interval (PK sampling): 60 +/− 30 minutes following the end of infusion (PK1), 210 +/− 30 minutes following the end of infusion (PK2), and <30 minutes prior to the next dose (PK3). All vancomycin concentrations were measured using the chemiluminescent microparticle immunoassay (Abbott Architect i2000SR, Abbott Diagnostics, Abbott Park, IL, USA) in the CHOP Chemistry Laboratory, a CLIA- and CAP-certified clinical laboratory. The analytical range of this assay was 3.0–50.0 mg/L; quality control (QC) was performed daily using Liquichek Immunoassay Plus Control (Bio-Rad, Hercules CA, USA)(assayed, tri-level).

Plasma CysC was measured using a residual plasma sample drawn clinically within 24 hours prior to the start of IV vancomycin infusion to provide a baseline value. Samples that were intentionally drawn for creatinine measurement for clinical care were utilized. In the event that there was no residual sample available, the baseline CysC level was obtained using the first clinical sample drawn following enrollment. The second sample was collected for CysC measurement along with PK3 as described above. CysC was measured using solid phase colorimetric sandwich ELISA (Quantikine ELISA, R&D Systems, Minneapolis, MN, USA) in the CHOP Translation Core Laboratory. Per the manufacturer, “This assay has been correlated to the cystatin C reference standard supplied by the Joint Research Centre Institute for Reference Materials and Measurements (Catalog # ERM-DA471/IFCC) with a slope of 1.07 and R2 value of 0.998.” The color intensity of each well was recorded by measuring their absorbance at 450 nm and 570 nm (SpectraMax 5, Molecular Devices, San Jose, CA, USA) consistent with the assay protocol. All measurements were performed in duplicate and were within the range of the assay (93.75–3000 ng/mL); three QC samples were run simultaneously and within their pre-specified ranges. Creatinine was measured by two-point rate spectrophotometric method (Vitros5600 analyzer, Ortho Clinical Diagnostics, Raritan, NJ, USA) in the CHOP Chemistry Laboratory; analytic range: 0.15–14.0 mg/dL. QC was performed daily using Liquichek Unassayed Chemistry Control (Bio-Rad, Hercules, CA, USA) (bi-level).

GFR was estimated using Schwartz_bed and various published CysC-based equations shown in Table 1.^27–32 In total, we calculated GFR using one creatinine-only equation (Schwartz_bed), five CysC-based equations, and one equation that combines creatinine and CysC values (full age spectrum [FAS_comb]). CysC values were rounded to the nearest 0.1 mg/L to be consistent with the number of significant digits of clinically reported creatinine values, which are rounded to the nearest 0.1 mg/dL by the CHOP Clinical Laboratory, and to be compatible with eGFR equations. The baseline CysC measurement was assumed to represent the CysC level in the entirety of the vancomycin course preceding the dose used for PK sampling, while the CysC level obtained coincident with PK sampling was used for the dosing interval during which PK sampling was performed. This approach, which relies on the most recent available measured value, was selected to mirror the clinical determination of renal function.

POPULATION PK MODEL DEVELOPMENT

Nonlinear mixed effects modeling was performed using NONMEM® software v7.4 and PDx-Pop v5.2.1 interface (ICON plc., Dublin, Ireland). First-order conditional estimation with interaction was used throughout model development. RStudio v1.1.456 (RStudio, Inc., Boston, MA) was used for graphical analysis. One- and two-compartment models were evaluated. Fixed effects were parameterized as V_d (volume of distribution) and CL (total body CL) for one-compartment models and CL, Q (intercompartmental CL), V1 (central V_d), and V2 (peripheral V_d) for two-compartment models. CL was allometrically scaled on the basis of bodyweight to 0.75 and V_d was scaled to 1 in the base model. Residual variability was evaluated using individual and combined additive in addition to proportional error models. Model selection was performed by evaluating goodness-of-fit diagnostic plots, comparing the minimum objective function value (OFV) and the Akaike information criterion (AIC), and assessing the precision of the estimates of the population fixed and random effect parameters. Because this study was conducted as an unfunded pilot study, formal power calculations were not used to determine the enrollment goal for the primary endpoint (population vancomycin CL).

Following the development of the base model, the selection of covariates for the model was conducted using a forward selection and then a backward stepwise elimination. This approach was used as the main objective of our study was to compare different estimates of GFR as a covariate of CL; forward selection would help directly compare model fit using different GFR estimating methods. A critical change in OFV of ≥3.84 was necessary for inclusion, and a threshold OFV increase of ≥7.88 was used for elimination from the final covariate model. All covariates were selected on the basis of physiologic plausibility. Dichotomous covariates (sex and the use of vasopressors and steroids) were included in the model to evaluate their effect on the respective PK parameter. Continuous covariates (age, height, creatinine level, cystatin C level, GFR estimated using different equations) were centered around the median population value and evaluated using the following power function:

$$P_i=TVP*{\left(\frac{CO{V}_i}{CO{V}_{med}}\right)}^{{\theta }_p}$$

where P_i is the estimated parameter for the ith subject, TVP is the typical value of the parameter, COV_i is the covariate value for a subject i, COV_med is the median value of the covariate for the study population, and θ_p is the covariate effect. Collinearity of covariates was considered when including eGFR since other measures are included in the calculation of eGFR (i.e. creatinine, cystatin C, height, gender).

Due to the small sample size of our study, we performed a leverage analysis to evaluate robustness of the parameter estimates. We examined the influence of individual subjects on the final model fit by removing each subject individually and refitting the model to the resulting datasets. Parameter estimates from each reduced dataset were recorded and compared with the final full model. A highly influential subject was defined as a subject whose removal resulted in a parameter estimate outside of the 95% confidence interval generated from the full dataset. We additionally calculated the percentage change of all parameter estimates from the full population estimate when subjects were removed.

SIMULATIONS

To visually evaluate the effect of the use of CysC compared with that of creatinine on estimates of vancomycin CL and drug concentrations, we performed Monte Carlo simulations (n = 500) using the baseline clinical characteristics of the study cohort (n = 20). We then implemented simulations using the final model based on Schwartz_bed at a dose of 15 mg/kg every 6 hours and repeated the simulations using the optimal CysC-based equation. The median simulated concentrations, along with 90% prediction intervals, were plotted for the fifth dose (i.e. dose given at 24 hours). Serum creatinine and CysC measurements were assumed to be stable over the simulated courses. Each patient’s median simulated CL estimates obtained by the two GFR estimating methods were compared according to the following equation: median CL_CysC - CL_Schwartz / CL_CysC × 100%.

Results

STUDY POPULATION

A total of 20 patients were included in the study and 83 vancomycin concentrations were utilized in the development of the population PK model, of which 58 were research and 25 were clinical samples. One vancomycin trough level drawn by the clinical team following a subject’s second vancomycin dose was below the level of quantification and was not included in the modeling process. The characteristics of the study population are shown in Table 2. Subjects’ age ranged from 3.9 to 18.2 years: seven were less than 10 years, seven were 10–14 years, and six were 15 years of age or older.

Table 2.

Characteristics of the study population

Characteristics	n = 20
Sex, female (%)	6 (30)
Age in years, median (interquartile range, IQR)	12.7 (8.8–15.5)
Weight (kg), median (IQR)	33.9 (25.9–45.9)
Height (cm), median (IQR)	115 (136–152)
Underlying disease, n (%)
Chronic lung disease	8 (40)
Neurologic disorder	8 (40)
Immunocompromising conditiona	5 (25)
Indication for vancomycin, n (%)
Sepsis/rule-out sepsis	13 (65)
Pneumonia	4 (20)
Otherb	3 (15)
Mechanical ventilation, n (%)	11 (55)
Initial vancomycin dosing (mg/kg/day), median (IQR)	54.8 (37.4–59.4)
Initial dosing frequency, n (%)
Every 6 hours	14 (70)
Every 8 hours	6 (30)
The use of steroids, n (%)	8 (40)
Baseline creatinine (mg/dL), median (IQR)	0.4 (0.3–0.7)
Baseline cystatin C (mg/L), median (IQR)c	0.6 (0.5–0.8)
Creatinine on PK sampling day (mg/dL), median (IQR)	0.4 (0.3–0.5)
Cystatin C on PK sampling day (mg/L), median (IQR)	0.5 (0.4–0.8)
eGFR at time of PK sampling (mL/min/1.73 m2), median (IQR)
Creatinine-based equation
Schwartzbed16	138 (97–179)
Cystatin C-based equations
Schwartzcysc29	114 (87–151)
CKD-EPI32	147 (123–167)
Hoek et al.28	130 (96–177)
Grubb et al.31	163 (119–226)
Zappitelli et al.30	138 (99–197)
Creatinine and cystatin C-based equation
Full age spectrum (FAScomb)27	145 (109–175)

^aIncludes two children with a history of hematopoietic stem cell transplant, one with Crohn’s disease, one with adrenal insufficiency, and one with X-linked agammaglobulinemia.

^bIncludes one indication each for bacteremia, skin/soft tissue structure infection, and postoperative administration.

^c15 children had cystatin C data available from residual sample prior to the initiation of vancomycin and five had cystatin C data drawn following enrollment.

Baseline CysC was measured using a residual plasma sample withdrawn before initiating vancomycin treatment in 15 subjects (75%) and following enrollment in 5 subjects (median 28.8 hours after the start of vancomycin infusion). A second CysC sample could not be withdrawn from one subject during PK sampling; thus, the same CysC value was used as both the baseline and PK sampling measurement (collected 7.8 hours prior to PK sampling and 88 hours after the start of vancomycin infusion). Two subjects met AKI criteria, defined as a 50% change in either creatinine or CysC level between vancomycin treatment initiation and PK sampling: one subject met the criteria owing to creatinine changes only and the other owing to CysC changes only.

POPULATION PK MODEL

A one-compartment model fit the observed data better than a two-compartment model, based on diagnostic plots and model performance. Supplemental Figure 1 displays the concentration-time plot of data used for the population PK analysis. The residual error was described using an additive plus proportional error structure. To test the appropriateness of using a fixed exponent of 0.75 when allometrically scaling CL on the basis of weight, we tested the inclusion of the exponent as a random effect parameter. The point estimate of this parameter was 0.74, thus we included 0.75 as a fixed value for the remainder of the model building process.

During the covariate selection process (Table 3), inclusion of eGFR as a covariate for CL provided greater initial OFV reduction than the base model, regardless of which GFR estimating equation was used. Each of the CysC-based eGFR equations tested resulted in a significantly greater reduction in OFV (ΔOFV range: −17.191 to −18.704) than Schwartz_bed (ΔOFV −12.820); the reduction in OFV for each of the CysC-based equations was comparable to the reduction in OFV when CysC was included as a covariate directly (−18.289). The FAS_comb equation provided the greatest initial reduction in OFV during the model building process (ΔOFV −22.913), which was significantly greater than that of Schwartz_bed or the CysC-based equations.

Table 3.

Stepwise covariate model building

Step	Minimization and covariance successful	OFV (Δ)	AIC (Δ)
1. Base model and error structure
One-compartment model with additive and proportional error	Yes	311.797	323.797
2. Base, covariate buildinga
Creatinine as a covariate on CL	Yes	302.864 (−8.933)	316.864 (−6.933)
Cystatin C as a covariate on CL	Yes	293.508 (−18.289)	307.508 (−16.289)
eGFR (Schwartzbed) as a covariate on CL	Yes	298.977 (−12.820)	312.977 (−10.820)
eGFR (Schwartzcysc) as a covariate on CL	Yes	293.395 (−18.402)	307.395 (−16.402)
eGFR (CKD-EPI) as a covariate on CL	Yes	294.606 (−17.191)	308.606 (−15.191)
eGFR (Hoek) as a covariate on CL	Yes	293.093 (−18.704)	307.093 (−16.704)
eGFR (Grubb) as a covariate on CL	Yes	293.704 (−18.093)	307.588 (−16.093)
eGFR (Zappitelli) as a covariate on CL	Yes	293.588 (−18.209)	307.588 (−16.209)
eGFR (FAScomb) as a covariate on CLb	Yes	288.884 (−22.913)	302.884 (−20.913)
3. Covariate building, with eGFR (FAScomb) on CL
Female sex as a covariate on CL	Yes	281.973 (−6.911)	297.979 (−4.911)

^aHeight, age, the use of vasoactive medications, and female sex were tested as covariates to determine their effect on CL and volume at each step. The use of steroids was tested as a covariate on CL at each step. Results not displayed as these parameters did not significantly improve model fit compared with the covariates shown.

^beGFR (FAS_comb) as a covariate on CL provided the greatest reduction in OFV and AIC during step 2 of model building and was utilized for the remaining covariate evaluation process. Other eGFR equations are shown as references.

CysC, cystatin C; eGFR, estimated glomerular filtration rate; OFV, objective function value; AIC, Akaike information criterion; CL, clearance; FAS_comb, full age spectrum

The final model included eGFR (from FAS_comb) and female sex as covariates on CL (Table 4). The diagnostic plots (Figure 1) demonstrate good fit between observed and predicted concentrations. Leverage analysis did not identify any highly influential subjects and all parameter estimates from the reduced datasets were well within the 95% confidence interval generated from the full model. The range of CL estimates with each subject removed was from 3.84 to 4.08 L/hr for a 34 kg patient. The range of V_d estimates with each subject removed was from 15.7 to 17.0 L for a 34 kg patient. The range of FAS_comb estimates was from 0.754 to 1.04.

Table 4.

Parameter estimates from final vancomycin population pharmacokinetic model

Parameter	Point estimate	95% Confidence Interval	Relative standard error (%)
CL,a L/hr (L/hr/kg0.75)	4.02 (0.29)	3.6–4.2 (0.26–0.30)	5.93
V,a L (L/kg)	16.4 (0.48)	13.8–19.0 (0.41–0.56)	8.05
eGFR from FAScomb	0.903	0.48–1.33	24.1
Female sex	0.699	0.48–0.92	16.0
Inter-individual variability,b
Clearance	23.4		34.9
Volume	NA		NA
Residual variability,
Proportionalc	0.179		65.1
Additive	2.44		106.0

^aParameter estimates presented for a typical 34-kg male patient with an eGFR of 145 mL/min/1.73 m². Final model:

CL (L/hr) = 4.02 × (WT/34)^0.75 × (eGFR/145)^.903 × (0.699 if female)

V (L) = 16.4 × (WT/34)

^bInter-individual variability point estimates are presented as percent coefficient of variation calculated by the square root of the variance × 100.

^cProportional residual variability reported as standard deviation.

CL, clearance; eGFR, estimated glomerular filtration rate; FAS_comb, full age spectrum; NA, not applicable; V, volume

Figure 1.

Diagnostic plots for the final population pharmacokinetic model. Top left: Observed vs. predicted population concentrations. Top right: Observed vs. predicted individual concentrations. Bottom left: Conditional weighted residuals vs. population predicted concentrations. Bottom right: Conditional weighted residuals vs. time after drug administration. Hashed blue lines represent the LOWESS fitted line of plotted data.

Replacement of eGFR using FAS_comb with Schwartz_bed in the final model led to a significant increase in the model’s OFV (+10.755) and an increase in inter-individual variability of the CL estimate (percent coefficient of variation [CV%] from 23.4% to 31.6%). The optimal CysC-based equation (Hoek et al.²⁸) also led to a significant increase in OFV (+5.706) compared with the FAS_comb model but a significant reduction in OFV compared with Schwartz_bed (−5.049).

The typical value of CL for the final model developed using FAS_comb was 5.1% lower than the typical value of CL using the final model with Schwartz_bed (4.02 vs. 4.23 L/hr for a patient weighing 34 kg). Plots of the Monte Carlo simulations are shown in Supplemental Figure 2. The concentration-time profiles across all simulations were similar when using eGFR calculated by both FAS_comb or Schwartz_bed in the model at a dose of 15 mg/kg every 6 hours, but the median C_min concentrations at 36 hours differed by >10% between FAS_comb or Schwartz_bed simulations: 12.8 vs. 11.3 mg/L. Additionally, there were significant differences in CL estimates at the individual subject level within simulations. Of the 20 simulated subjects (n = 500 simulations per subject), median CL estimates using the two equations differed by >10% in half of the subjects, and the average difference in median CL estimates was −10.2% (range: −46.6 to 12.3%).

Discussion

In our study of critically ill children prescribed IV vancomycin, incorporation of CysC-based eGFR as a covariate of CL significantly improved model fit compared to the use of the most frequently used creatinine-based equation, the Schwartz_bed equation. This finding was consistent across multiple different CysC-based equations,^28–32 including an equation combining creatinine and CysC,²⁷ suggesting that CysC outperforms creatinine as an estimate of renal function in critically ill children. The equation that best described vancomycin CL in our study population, FAS_comb, includes both creatinine and CysC to estimate GFR and was found to be superior to all creatinine and CysC-based equations that were evaluated. Routine measurement of CysC and creatinine may best indicate the optimal dosage of vancomycin in individual patients, while CysC alone improves estimation of vancomycin CL compared to the creatinine-based approach.

Critically ill children demonstrate significant physiologic variability owing to fluid shifts, hemodynamic changes, and medication administration. This can lead to important alterations in drug disposition and CL. Vancomycin is a particularly important medication in the critical care setting for which early, targeted exposures are necessary to ensure optimal outcomes.⁴ Vancomycin is eliminated almost exclusively via glomerular filtration, but creatinine-based eGFR equations do not accurately predict vancomycin levels in children.¹⁵ According to our findings, measurement of CysC can improve vancomycin CL estimation in critically ill children compared to the measurement of creatinine alone: the typical value of CL was 5% lower when using a combined CysC- and creatinine-based eGFR equation, FAS_comb, compared with that using Schwartz_bed, and simulated trough concentrations were >10% higher than those with Schwartz_bed. Concerns regarding nephrotoxicity and delayed changes in creatinine often limit clinicians’ willingness to administer higher vancomycin dosages (60–80 mg/kg/day) needed in most children to achieve therapeutic targets (AUC 400–600 mg*hr/L). Measurement of CysC may provide more reliable estimates of renal function, especially vancomycin CL; decrease the likelihood of perpetuating toxicity in the setting of AKI; and promote personalized, target-oriented dosing.

Most GFR estimating equations have been developed to describe kidney function in children with impaired chronic kidney disease.^16,29,34 This is, in part, because of the difficulty in directly measuring GFR in healthy children.³⁵ From a pharmacologic perspective, determining impaired kidney function is crucial to avoid overdose of renally excreted drugs. Irrespective, when administering antibiotics in critically ill children, it may be equally important to accurately identify normal or augmented renal CL to ensure sufficient drug exposure to eradicate bacterial infections. The renal function in our study population was largely normal. Four patients (20%) had eGFR ≤ 120 ml/min/1.73 m² according to Schwartz_bed at the time of PK sampling, and one patient had eGFR ≤ 90 ml/min/1.73 m². Interestingly, all of these children were found to have eGFR ≤ 120 ml/min/1.73 m² accordign to FAS_comb, but three of them had eGFR ≤ 90 ml/min/1.73 m². In our study, the use of FAS_comb provided a better estimation of kidney function considering vancomycin CL. Additional studies are needed to better compare creatinine and CysC-based equations for estimating the CL of other drugs in critically ill children, as well as vancomycin CL among children with moderate-to-severe kidney function impairment.

Recently published guidelines call for AUC-directed therapy in all patients,³⁶ including children, and advocate the use of Bayesian dosing software to estimate vancomycin AUC. As a result, accurate assessment of renal function is imperative. Most available population PK models of vancomycin in children are derived from non-critically ill populations and utilize creatinine to estimate GFR. But, as we have demonstrated, measurement of CysC promotes better estimation of vancomycin CL than creatinine alone, which can facilitate safer administration of this antibiotic. Our findings are consistent with a recently published population PK model by Lu et al.,²⁶ which suggested CysC as an informative covariate of vancomycin CL among 220 hospitalized children. Our study extends this finding to a critical care population specifically and demonstrates that CysC is more informative than creatinine for vancomycin CL estimation.

While FAS_comb was found to best describe vancomycin CL, it is noteworthy that each of the CysC-based equations led to a significantly larger reduction in OFV than Schwartz_bed in our model development process. Moreover, each of the CysC-based equations led to similar reductions in OFV compared to one another (Table 3), supporting the interchangeability of these equations for estimating GFR.³⁵ These equations outperformed Schwartz_bed despite the measurement of CysC at only two time points during the study, while creatinine was measured at least daily. This further supports the superiority of CysC over creatinine for estimating CL of vancomycin in children.

In our final population model, female sex was associated with lower vancomycin CL. To our knowledge, differences attributable to sex in vancomycin CL have not been previously reported. While this may be the result of the small sample size in our study, this is also a plausible finding. Creatinine is produced at a lower rate in females than in males, and many eGFR equations account for this difference when estimating GFR or creatinine CL. FAS_comb and Schwartz_bed do not specifically adjust for sex, and therefore sex may be an important component when estimating vancomycin CL using these two equations (ΔAIC when adding female sex to Schwartz_bed during model building was −4.249, data not shown). Additional studies are needed to further explore this association.

There are notable limitations to our study. First, this was a small study conducted in only 20 subjects. It was not designed to specifically include children over the full range of possible renal function values. Additional studies are required to characterize the PK of vancomycin in children with moderate-to-severe renal dysfunction. Similarly, this study was designed to specifically compare Schwartz_bed and CysC-based eGFR equations as covariates on CL. We did not investigate all possible covariate effects that could be explored in a critically ill population (i.e. fluid volume status, severity of illness indicators, etc.). Additionally, CysC was measured at only two time points during the study, which may have affected the precision of CysC-based renal function estimates. However, all patients had CysC measured during the time frame when PK sampling was performed, which is likely the most informative and most important time to have accurate renal function estimation. Finally, CysC assays are not routinely clinically available at all institutions and measurements can vary based on the method and instrument used,³⁷ which may influence the generalization of our findings. We measured CysC in a research lab using a validated approach, but CysC-based eGFR estimates can differ based on measurement approaches, and clinicians should be aware of this when utilizing CysC for dosing recommendations.

Conclusion

Thus, measurement of CysC improved vancomycin CL estimation in our population PK model of critically ill children. Whether measured alone or in combination with creatinine, CysC can help estimate GFR that more reliably correlates with vancomycin CL than the most commonly used creatinine-based equation in children, Schwartz_bed. We believe that our findings are important and add to the existing body of literature demonstrating the superiority of CysC over creatinine for estimating drug concentrations and CL.³⁸ Larger studies involving children with varying renal function should be conducted to identify the ability of CysC to promote individualized dosing of vancomycin in all critically ill children.

Supplementary Material

supp fig 1

Supplemental Figure 1. Concentration-time plot of the data used for establishing a population pharmacokinetic model of vancomycin in critically ill children. Hashed line represents the LOWESS fitted line of plotted data.

supp fig 2

Supplemental Figure 2. Concentration-time profiles for simulated subjects using bedside Schwartz (blue) and full age spectrum (red) equations on vancomycin clearance. Solid lines: median concentrations. Hashed lines: 90% prediction interval (fifth and ninety-fifth percentile of simulated concentrations).