The most appropriate vancomycin dosing strategy in pediatric patients weighing ≥70 kg (weight based versus non-weight based) to achieve an area under the concentration-time curve (AUC) of ≥400 mg·liter/h and a trough concentration of <20 mg/liter is not known. Population pharmacokinetic analysis determined that dosing of vancomycin should be weight based using fat-free mass, with appropriate adjustment for kidney dysfunction.
KEYWORDS: obesity, pediatrics, population pharmacokinetics, vancomycin
ABSTRACT
The most appropriate vancomycin dosing strategy in pediatric patients weighing ≥70 kg (weight based versus non-weight based) to achieve an area under the concentration-time curve (AUC) of ≥400 mg·liter/h and a trough concentration of <20 mg/liter is not known. Population pharmacokinetic analysis determined that dosing of vancomycin should be weight based using fat-free mass, with appropriate adjustment for kidney dysfunction.
TEXT
Current guidance for dosing vancomycin in pediatric patients, to achieve an area under the concentration-time curve from 0 to 24 h (AUC0–24)/MIC ratio of ≥400, is 10 to 15 mg/kg of body weight/dose every 6 to 8 h in patients with normal kidney function, whereas in adults standard doses of 1 to 2 g every 6 to 12 h are often used (1). A weight (WT) of 70 kg has often been used as a marker for using adult dosing, as this is when a dose of 15 mg/kg/dose is above the adult starting dose of 1,000 mg. The optimal method for vancomycin dosing in pediatric patients of ≥70 kg is unknown.
We performed a retrospective pharmacokinetic modeling and simulation study determining the appropriate weight and methodology for dosing (milligram per kilogram per dose or milligram per dose) of vancomycin in pediatric patients with a body weight of ≥70 kg.
The Texas Children's Hospital electronic medical record was queried from 1 January 2011 to 30 June 2018 to include patients who had received intravenous vancomycin, had a baseline serum creatinine (SCR) within 24 h of vancomycin initiation, were <19 years of age, and were greater than or equal to 70 kg body weight. Patients were excluded if they were undergoing renal replacement therapy while receiving vancomycin or mechanical circulatory support. Data collected from 1 January 2011 to 30 June 2017 were used for model building and simulation, and data from 1 July 2017 to 30 June 2018 were used as a validation data set.
Data collection included WT (kilograms), height (centimeters), gender, age (years), SCR (milligrams per decaliter), urine output over the prior 12 h, vancomycin dose (milligrams), and vancomycin serum concentrations (milligrams per liter). Dosing was per the discretion and clinical judgement of the provider. Calculated variables included fat-free mass (FFM), estimated creatinine clearance (CrCl), body mass index (BMI), and BMI percentile for age and gender (2, 3) (Appendix). The percentage of the study population who were obese (≥95th percentile for age and gender) or overweight (85 to 94th percentile for age and gender) was also determined (4). The MIC was considered to be 1 mg/liter.
Descriptive analysis using percent, means, standard deviations, medians, and interquartile ranges (IQR) was performed using Stata IC v.12 (StataCorp LLC, College Station, TX) and Excel 2013 (Microsoft Corp, Redmond, WA). Population pharmacokinetic analysis was performed with NONMEM v.7.3 (Icon, PLC, Dublin, Ireland) and PDx-Pop 5.2 (Icon, PLC, Dublin, Ireland) using first-order conditional estimation with interaction (FOCE-I). One- and two-compartment models were evaluated, and interindividual variability (IIV) was modeled exponentially.
Potentially significant covariates, based on clinical utility, were initially plotted against pharmacokinetic parameters and η-plots. Covariates were retained in the model if the reduction in the objective function value (OFV) was >3.84, signifying statistical significance (P < 0.05) at one degree of freedom. After a full model was developed with all significant covariates, individual covariates were then removed from the model and considered to be retained in the model if the OFV increased by >10.83 (5, 6).
To evaluate model goodness of fit and detect bias, scatter plots of dependent variables (DV) versus predicted (PRED) and individual predicted (IPRE) values were developed, as were conditional weighted residuals (CWRES) versus PRED and CWRES versus time after dose (7) (Fig. 1). Bootstrap simulations were performed (n = 1,000) on the final model with calculation of 95% confidence intervals.
FIG 1.
Goodness-of-fit graphs. The graphs do not demonstrate any type of bias and show a good fit of the final model.
A simulation data set was developed based on the demographics and covariates from the initial data set. Using the developed pharmacokinetic model, each subject in the data set was simulated 1,000 times with trough concentrations simulated prior to the 4th dose. Endpoints for the simulation included AUC0–24/MIC ratio (means and standard deviations), percentage of dosing regimens to achieve an AUC0–24/MIC ratio of ≥400, trough concentrations (means and standard deviations), and percentage of trough concentrations of <20 mg/liter.
A validation data set was developed from patients (1 July 2017 to 30 June 2018) who met original study criteria. Median prediction error and absolute prediction error with 95% confidence intervals for serum concentrations were generated during a simulation process using the model IPRE compared to the actual concentrations using the MAXEVAL = 0 POSTHOC command in NONMEM (Appendix). Graphical representation of normalized prediction distribution error (NPDE) was developed.
Initially, 626 patients were identified (1 January 2011 to 30 June 2017), and 196 patients met study criteria (Table 1). Included patients had a total of 8 doses (IQR, 5 to 13) infused over 1 h at a mean dose of 1,192 ± 205 mg (13.3 ± 2.2 mg/kg/dose by total body weight, 20.9 ± 3.8 mg/kg/dose by FFM). A total of 555 serum vancomycin concentrations were included (57 concentrations [10.3%] were below quantification of the assay [median time after dose, 8.5 h; range, 5.9 to 17.4 h] and were excluded). Patients had a median of 1 (IQR, 1 to 3) serum vancomycin concentration sampled at a median of 7.9 h (range, 0.2 to 112.4 h) after a dose, and the mean serum concentration was 12.0 ± 5.4 mg/liter.
TABLE 1.
Patient demographics
| Parameter (n = 196) | Value |
|---|---|
| Male (%) | 68.9 |
| Age (yr) (median, range) | 15.9 (9.3, 18.9) |
| Serum creatinine (mg/dl) (mean, standard deviation) | 0.90 ± 0.48 |
| Creatinine CL (ml/min/1.73 m2) | 100 ± 36 |
| Weight (kg) (mean, standard deviation) | 91.8 ± 20.6 |
| FFM (kg) (mean, standard deviation) | 57.9 ± 10.0 |
| Height (cm) (mean, standard deviation) | 169 ± 10 |
| BMI (kg/m2) (mean, standard deviation) | 32.4 ± 6.8 |
| BMI (%) (mean, standard deviation) | 94.8 ± 8.6 |
| Obese (%) | 75.0 |
| Overweight (%) | 13.8 |
| Urine output over the prior 12 h (ml) (mean, standard deviation) | 899 ± 743 |
| Urine output over the prior 12 h (ml/kg/h) (mean, standard deviation) | 0.83 ± 0.71 |
A one-compartment proportional error model best fit the data. WT and FFM were applied to clearance (CL) and volume of distribution (V) with allometric scaling (0.75 on CL, 1 on V). The decrease in OFV when WT was used was not significant from the model with no weight (−1.708), but the decrease in OFV was greater and statistically significant with the FFM model (−10.787). FFM was used for further covariate modeling. SCR was placed on CL exponentially and resulted in a significant decrease in OFV (−274.857) and improved overall model fit. Neither age nor urine output reached statistical significance when placed in the model (Table 2). Diagnostic plots showed good fit of the model with minimal bias (Fig. 1). The final η values for CL (13.6%) and for V (47.5%) were similar to the initial base model values of 8.9% and 41.8%, respectively. Bootstrapping with 1,000 replications was performed, and 97.8% of runs were successful (Table 3).
TABLE 2.
Final model and variabilitya
| Model (n = 196) | IIV (%) | Residual variability (%) |
|---|---|---|
| CL = 18.6 × (FFM/70)0.75 × 0.582(SCR/0.67) | 32.6 | 21.7 |
| V = 102 × (FFM/70) | 40.5 |
CL, clearance (liters/h); FFM, fat-free mass in kilograms; SCR, serum creatinine (mg/dl); V, volume of distribution (liters); IIV, interindividual variability.
TABLE 3.
Bootstrap analysisa
| Parameter (n = 196) | Value(s) for: |
|||
|---|---|---|---|---|
| 1-Compartment model |
1,000 Bootstrap |
|||
| Estimate (RSE%) | 95% CI | Median (RSE%) | 95% CI | |
| CL (liters/h) | 18.6 (5.5) | 16.6–20.6 | 18.7 (6.9) | 16.4–21.2 |
| V (liters) | 102 (10.1) | 81.8–122.2 | 103 (11.9) | 84.1–129 |
| Serum creatinine (mg/dl) | 0.58 (3.6) | 0.54–0.62 | 0.58 (4.6) | 0.52–0.62 |
| ω1 (%) | 32.6 (15.8) | 27.0–37.3 | 32.2 (24.9) | 26.6–37.4 |
| ω2 (%) | 40.5 (33.3) | 23.8–52.1 | 40.5 (35.4) | 25.4–53.6 |
| Proportional error (%) | 21.7 (9.7) | 19.5–23.7 | 21.6 (9.9) | 19.3–23.6 |
RSE%, relative standard error; 95% CI, 95% confidence interval; ω1, interindividual error on CL; ω2, interindividual error on V.
The simulation data set, composed of the original 626 patients, using CL- and V-generated estimates to simulate steady-state exposure levels (trough concentrations prior to the 4th dose), identified that a dose of 20 mg/kg/dose every 6 h based on FFM achieved the greatest percentage of simulated subjects with an AUC0–24/MIC ratio of ≥400 and a trough concentration of <20 mg/liter (Table 4 and Fig. 2).
TABLE 4.
Dosing simulation based upon fat-free mass
| Dosea (n = 626) | AUC (mg·h/liter) (mean, standard deviation) | AUC0–24/MIC of >400 (%) | Trough (mg/liter) (mean, standard deviation) | Trough below 20 mg/liter (%) |
|---|---|---|---|---|
| 10 mg/kg q8 (569.79 ± 107.82 mg) | 191 ± 40 | 0 | 5.3 ± 1.2 | 100 |
| 10 mg/kg q6 (569.79 + 107.82 mg) | 255 ± 53 | 2 | 7.3 ± 1.4 | 100 |
| 15 mg/kg q8 (854.69 ± 161.73 mg) | 287 ± 59 | 5 | 7.9 ± 1.9 | 100 |
| 15 mg/kg q6 (854.69 ± 161.73 mg) | 383 ± 79 | 32 | 10.9 ± 2.1 | 100 |
| 20 mg/kg q8 (1,139.58 ± 215.64 mg) | 383 ± 79 | 32 | 10.6 ± 2.5 | 100 |
| 20 mg/kg q6 (1,139.58 ± 215.64 mg) | 510 ± 105 | 91 | 14.6 ± 2.7 | 95 |
| 25 mg/kg q8 (1,424.48 ± 269.55 mg) | 478 ± 99 | 78 | 13.3 ± 3.1 | 96 |
| 25 mg/kg q12 (1,424.48 ± 269.55 mg) | 319 ± 66 | 11 | 7.6 ± 2.3 | 100 |
| 25 mg/kg q24 (1,424.48 ± 269.55 mg) | 159 ± 33 | 0 | 2.0 ± 1.0 | 100 |
| 30 mg/kg q8 (1,709.37 ± 323.46 mg) | 574 ± 119 | 99 | 15.9 ± 3.7 | 86 |
| 30 mg/kg q12 (1,709.37 ± 323.46 mg) | 383 ± 79 | 32 | 9.1 ± 2.8 | 100 |
| 30 mg/kg q24 (1,709.37 ± 323.46 mg) | 191 ± 40 | 0 | 2.4 ± 1.2 | 100 |
| 500 mg q8 (9.07 ± 1.58 mg/kg) | 171 ± 35 | 0 | 4.8 ± 1.1 | 100 |
| 500 mg q6 (9.07 ± 1.58 mg/kg) | 229 ± 46 | 1 | 6.6 ± 1.2 | 100 |
| 1,000 mg q8 (18.13 ± 3.16 mg/kg) | 343 ± 69 | 15 | 9.5 ± 2.1 | 100 |
| 1,000 mg q6 (18.13 ± 3.16 mg/kg) | 457 ± 93 | 75 | 13.1 ± 2.5 | 98 |
| 1,250 mg q8 (22.67 ± 3.95 mg/kg) | 429 ± 87 | 60 | 11.9 ± 2.6 | 99 |
| 1,250 mg q6 (22.67 ± 3.95 mg/kg) | 571 ± 116 | 98 | 16.4 ± 3.1 | 91 |
| 1,500 mg q8 (27.2 ± 4.74 mg/kg) | 514 ± 104 | 92 | 14.3 ± 3.1 | 94 |
| 1,500 mg q6 (27.2 ± 4.74 mg/kg) | 685 ± 139 | 100 | 19.7 ± 3.7 | 61 |
q6, q8, q12, and q24 indicate every 6, 8, 12, and 24 h, respectively.
FIG 2.
Simulated attainment of goal AUC/MIC ratios and serum concentrations of <20 mg/liter with various dosing regimens and degrees of kidney dysfunction. Dosing of 20 mg/kg/dose every 6 h most often meets goal AUC/MIC ratios and serum concentrations with reductions in dosing interval to every 8 h for moderate kidney dysfunction.
A validation data set of 50 patients (16.1 ± 2.2 years, WT of 91.1 ± 18.0 kg, 74.0% male) met original study criteria (FFM, 58.4 ± 10.2 kg; SCR, 0.89 ± 0.43 mg/dl). Obesity was present in 66.0% of patients (n = 33). Mean vancomycin dose was 1,175 ± 210 mg (13.4 ± 2.5 mg/kg/dose by WT, 20.8 ± 4.9 mg/kg/dose by FFM). Vancomycin serum concentrations (n = 119; 14.6 ± 10.8 mg/liter) were sampled at a median of 10.7 h (IQR 7.6 to 13.0 h) after a dose. Median individual prediction error was −4.2% (95% confidence interval, −4.4% to 3.6%), and median absolute individual prediction error was 10.2% (10.8% to 17.1%) with normal distribution of errors (Fig. 3).
FIG 3.
Normalized prediction errors for the validation data set. A lack of bias is noted.
The identification of FFM as a better descriptor of vancomycin pharmacokinetics than WT is unique to this report, as prior investigations in pediatric patients have demonstrated that using total body weight in obese pediatric patients is acceptable for dosing of vancomycin (8–10). There is potential for supratherapeutic or subtherapeutic dosing if FFM is not used, and we feel strongly that these calculations should be implemented into clinical practice.
It is well documented that vancomycin is primarily renally eliminated, and adjustments in dose should occur for diminished kidney function. Prior publications have noted that age is a significant covariate for vancomycin pharmacokinetic disposition in the pediatric population, most likely due to extended courses of vancomycin and subsequent increases in SCR, but we did not identify this as a significant covariate (1, 11).
Prospective evaluation of the pharmacokinetic model and dosing recommendations, ideally through Bayesian analysis, would allow for modifications to the current model and identify shortcomings or other currently unidentified potentially significant covariates. We recommend the monitoring of serum vancomycin concentrations in this population in order to ensure attainment of goal trough concentrations and AUC0–24/MIC ratios.
Vancomycin pharmacokinetics are best described by fat-free mass and serum creatinine in pediatric patients with an actual body weight of ≥70 kg. Vancomycin dosing in this patient population should be based on fat-free mass, with adjustments in dose for diminished kidney function.
APPENDIX
For fat-free mass calculation, the equations used were FFMfemales = [1.11 + ((1 – 1.11)/(1 + (age/7.1)−1.1))] × [(9,270 × WT)/(8,780 + 244 × BMI)] and FFMmales = [0.88 + ((1 – 0.88)/(1 + (age/13.4)−12.7))] × [(9,270 × WT)/(6,680 + 216 × BMI)]. BMI calculation used the equation BMI = kg/(HT × 100)2. Equations for prediction error calculations were the following: prediction error = (concentrationIPRE – concentrationobserved)/concentrationobserved and absolute prediction error = |concentrationIPRE – concentrationobserved|/concentrationobserved. The following abbreviations and/or units of measurment were used for these equations: HT, height (cm); age, in years; WT, actual body weight (kg); BMI, body mass index (kg/m2); FFM, fat-free mass (kg); IPRE, individual predicted values.
REFERENCES
- 1.Le J, Bradley JS, Murray W, Romanowski GL, Tran TT, Nguyen N, Cho S, Natale S, Bui I, Tran TM, Capparelli EV. 2013. Improved vancomycin dosing in children using area under the curve exposure. Pediatr Infect Dis J 32:e155–e163. doi: 10.1097/INF.0b013e318286378e. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Al-Sallami HS, Goulding A, Grant A, Taylor R, Holford N, Duffull SB. 2015. Prediction of fat-free mass in children. Clin Pharmacokinet 54:1169–1178. doi: 10.1007/s40262-015-0277-z. [DOI] [PubMed] [Google Scholar]
- 3.Schwartz GJ, Munoz A, Schneider MF, Mak RH, Kaskel F, Warady BA, Furth SL. 2009. New equations to estimate GFR in children with CKD. J Am Soc Nephrol 20:629–637. doi: 10.1681/ASN.2008030287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ogden CL, Carroll MD, Lawman HG, Fryar CD, Kruszon-Moran D, Kit BK, Flegal KM. 2016. Trends in obesity prevalence among children and adolescents in the United States, 1988-1994 through 2013-2014. JAMA 315:2292–2299. doi: 10.1001/jama.2016.6361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Beal SL. 2002. Commentary on significance levels for covariate effects in NONMEM. J Pharmacokinet Pharmacodyn 29:403–410. doi: 10.1023/A:1020909324909. [DOI] [PubMed] [Google Scholar]
- 6.Wahlby U, Jonsson EN, Karlsson MO. 2002. Comparison of stepwise covariate model building strategies in population pharmacokinetic-pharmacodynamic analysis. AAPS PharmSci 4:E27. doi: 10.1208/ps040427. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Karlsson MO, Savic RM. 2007. Diagnosing model diagnostics. Clin Pharmacol Ther 82:17–20. doi: 10.1038/sj.clpt.6100241. [DOI] [PubMed] [Google Scholar]
- 8.Moffett BS, Kim S, Edwards MS. 2011. Vancomycin dosing in obese pediatric patients. Clin Pediatr (Phila) 50:442–446. doi: 10.1177/0009922810393500. [DOI] [PubMed] [Google Scholar]
- 9.Miller M, Miller JL, Hagemann TM, Harrison D, Chavez-Bueno S, Johnson PN. 2011. Vancomycin dosage in overweight and obese children. Am J Health Syst Pharm 68:2062–2068. doi: 10.2146/ajhp110107. [DOI] [PubMed] [Google Scholar]
- 10.Le J, Capparelli EV, Wahid U, Wu YS, Romanowski GL, Tran TM, Nguyen A, Bradley JS. 2015. Bayesian estimation of vancomycin pharmacokinetics in obese children: matched case-control study. Clin Ther 37:1340–1351. doi: 10.1016/j.clinthera.2015.05.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Knoderer CA, Gritzman AL, Nichols KR, Wilson AC. 2015. Late-occurring vancomycin-associated acute kidney injury in children receiving prolonged therapy. Ann Pharmacother 49:1113–1119. doi: 10.1177/1060028015594190. [DOI] [PubMed] [Google Scholar]