Population Pharmacokinetics and Dosage Optimization of Linezolid in Critically Ill Pediatric Patients

Linezolid is an oxazolidinone antibiotic exhibiting efficacy against multidrug-resistant (MDR) Gram-positive-related infections. However, its population pharmacokinetic (PopPK) profile in critically ill Chinese children has not been characterized.

ABSTRACT

Linezolid is an oxazolidinone antibiotic exhibiting efficacy against multidrug-resistant (MDR) Gram-positive-related infections. However, its population pharmacokinetic (PopPK) profile in critically ill Chinese children has not been characterized. Optimal dosing regimens should be established according to the population pharmacokinetic (PopPK)/pharmacodynamic (PD) properties of linezolid in the specific population. This work aims to describe the pharmacokinetic (PK) properties of linezolid, assess the factors affecting interpatient variability, and establish an optimized regimen for children in pediatric intensive care units (PICU). A single-center, prospective, open-labeled PK study was performed. Ultraperformance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) was applied to measure the plasma levels during linezolid treatment. PopPK analysis was conducted using Phoenix NLME software. A total of 63 critically ill pediatric patients were included. The data showed good fit for a two-compartment model with linear elimination. Body weight and aspartate aminotransferase (AST) were the most significant covariates explaining variabilities in linezolid PK for the pediatric population. The therapeutic target was defined as the ratio of the area under the drug plasma concentration-time curve over 24 h to a MIC (AUC/MIC) of >80. Different dosing regimens were evaluated using Monte Carlo simulation to determine the optimal dosage strategy for linezolid. Although the probability of target attainment (PTA) was high (>96%) for 10 mg/kg body weight every 8 h at a MIC of ≤1 mg/liter, it was lower than 70% at a MIC of >1 mg/liter. Thus, the dosing regimen required adjustment. When the dosing regimen was adjusted to 15 mg/kg every 6 h, the PTA increased from 63.6% to 94.6% at a MIC of 2 mg/liter, thereby indicating a higher degree of treatment success. Children with AST of >40 U/liter had a significantly higher AUC than those with AST of ≤40 U/liter (205.45 versus 159.96). Therefore, dosage adjustment was required according to the AST levels. The PopPK characteristics of linezolid in critically ill children were evaluated, and an optimal dosage regimen was constructed based on developmental PopPK/PD model and simulation. (This study has been registered in the Chinese Clinical Trial Registry under no. ChiCTR1900021386).

TEXT

Linezolid is an oxazolidinone antibiotic commonly used in pediatric intensive care units (PICU) to treat infections with multiresistant Gram-positive cocci (MR-GPC), especially methicillin-resistant Staphylococcus aureus (MRSA), and has been approved for pediatrics, including neonates. However, the pharmacokinetic (PK) profiles of Chinese pediatric patients with serious infections in PICU have not been fully characterized. Linezolid is primarily metabolized through morpholine ring oxidation, which results in two inactive ring-opened carboxylic acid metabolites: the aminoethoxyacetic acid metabolite (A) and the hydroxyethyl glycine metabolite (B). Nonrenal clearance accounts for approximately 65% of the total clearance of linezolid. Under steady-state conditions, approximately 30% of the dose appears in the urine as linezolid, 40% as metabolite B, and 10% as metabolite A (1).

Pharmacodynamic (PD) studies indicated that the therapeutic efficacy of linezolid is correlated with the ratio of the area under the drug plasma concentration-time curve over 24 h to the MIC (AUC/MIC) or percentage time above the MIC (%T>MIC) (2). A target AUC/MIC ratio of >80 (3, 4) is associated with pathogen eradication and good clinical outcome. Good clinical outcome is defined as the recovery of symptoms, signs, laboratory examination, and bacteriological tests or obvious improvement of the disease. In critically ill children, pathophysiology changes may affect the drug PK; thus, adjusting the dosing regimen should be inevitably considered. The most common serious adverse reaction caused by linezolid is hematological toxicity, such as thrombocytopenia (5, 6). Patients should be closely monitored for hematological indicators throughout linezolid treatment. The use of the optimal dosage in pediatric patients may improve treatment efficacy and decrease toxicity, including thrombocytopenia, anemia, leukopenia, and pancytopenia (7, 8).

This study aimed to develop a new population PK (PopPK) model. Various covariates related to the underlying conditions of critically ill pediatric patients were evaluated to identify patient-specific characteristics that can explain interindividual variability and to establish an evidence-based dosage regimen on the basis the developmental PopPK/PD model.

RESULTS

Study population.

A total of 63 critically ill patients with staphylococcal infections diagnosed from 2018 to 2019 with an age of 5.21 ± 4.22 years and weight of 22.28 ± 15.00 kg (mean ± standard deviation [SD]) were included. The number of children age 0 to 1 year, 1 to <12 years, and 12 to 18 years are 14, 43, and 6, respectively. Table 1 shows the summary of patient characteristics and demographics.

TABLE 1.

Baseline characteristics of 63 pediatric patients

Characteristica	Mean	SD	Median	Range or ratio
Age (yrs)	5.21	4.22	3.85	0.10–15.30
Gender	43 (M):20 (F)
Wt (kg)	22.28	15.00	15.00	4.20–70.00
Body mass index (kg/m2)	17.11	3.29	16.00	12.10–28.50
Linezolid dose (mg/kg/day)	28.85	2.96	30	18.46–31.03
Albumin (g/liter)	30.44	5.06	29.30	22.30–44.70
No. of samples (per patient)	4	2–4
Alanine aminotransferase (U/liter)	213.60	812.51	33.20	2.30–6258.40
Aspartate aminotransferase (U/liter)	300.65	1442.50	45.90	13.20–11320.90
Creatinine clearance (ml/min)	136.46	80.50	142.00	13.00– 554.00
Total bilirubin (μmol/liter)	22.92	11.23	11.23	1.55–190.01
Direct bilirubin (μmol/liter)	9.37	20.23	2.81	0.01–101.73
Pediatric critical illness score	16.71	11.20	11.20	0.20–93.60
CRRT/ECMO	15:2

^aCRRT, continuous renal replacement therapy; ECMO, extracorporeal membrane oxygenation. Creatinine clearance was calculated with the modified Schwartz equation.

Model building.

A total of 246 linezolid concentrations in 63 infected patients were included for the PopPK model. The concentration of linezolid samples ranged from 0.28 mg/liter to 41.19 mg/liter. The concentration-time curve is shown in Fig. 1.

FIG 1.

Observed concentration (DV) versus time after dose (TAD) profiles of linezolid. ((A) Semi-log plot; (B) linear plot).

A two-compartment model with first-order (linear) elimination was established to describe the concentration-time profiles of linezolid. The model had a lower objective function value (OFV) value (ΔOFV = 149.81, P < 0.001) and residual variability than a one-compartment model. The drop in residual variability between the two models was from 0.46 to 0.30. The central volume of distribution (V₁), peripheral volume of distribution (V₂), clearance (CL), and intercompartment clearance (CL₂) of linezolid were estimated in the present model. Interindividual variability was introduced as log-normal on different parameters, namely, V, CL, and CL₂, and then estimated. Residual variability was best described by a multiplicative error model.

Covariate analysis.

In the forward-inclusion and backward-elimination covariate stepwise search steps, the two most significant covariates were the positive influence of body weight on CL and CL₂ and the inverse association between AST and CL. The trend was also observed in age, which was correlated with weight (Fig. 2). None of the other evaluated covariates had any significant influence on the optimization of the model. Therefore, body weight and AST were included in the final model. The parameter estimates of the final linezolid PopPK model are presented in Table 2. The interindividual variation (CV%) of the estimation of CL and CL₂ decreased substantially, from 77.75% and 78.25% to 52.51% and 53.45%, respectively.

FIG 2.

Correlation plot between covariates age and weight (the blue line is the trend line).

TABLE 2.

Parameter estimates of final PopPK model and bootstrap results^a

Full data set		Bootstrap analysis (n = 1,000)
Parameter	Estimate	CV (%)	Median	95% CI	RSE%
tvV (liters)	5.22	12.47	5.23	3.50–6.53	0.19
tvV2 (liters)	28.79	13.11	29.29	20.90–42.39	1.74
tvCL (liters/h)	2.34	7.78	2.33	1.99–2.73	−0.43
tvCL2 (liters/h)	7.14	17.14	6.83	4.46–11.74	−4.34
dCLdWeight	0.80	12.36	0.81	0.61–1.03	1.25
dCL2dWeight	1.09	16.80	1.11	0.67–1.57	1.83
dCLdAST	–0.16	27.89	−0.16	−0.26 to −0.03	0.00
IIV_V (%)	55.78	56.12	52.62	2.25–91.05	−5.67
IIV_CL (%)	52.51	15.06	51.49	42.07–58.87	−1.94
IIV_CL2 (%)	53.45	43.17	53.29	0.13–79.70	−0.30
Residual variability	0.29	12.73	0.28	0.20–0.36	−3.45

^aThe final PopPK model was V (liters) = 5.22 · exp(ηV); V₂ (liters) = 28.79; CL (liters/h) = 2.34 · (weight/15) ^ 0.8 · (AST/45.9) ^ (–0.16) · exp(ηCL); CL₂ (liters/h) = 7.14 · (weight/15) ^ 1.09 · exp(ηCl₂). RSE%, (median of bootstrap – estimate)/estimate · 100; TV, typical population value; IIV, interindividual variability; 95% CI, 95% confidence interval; V, central volume of distribution; V₂, peripheral volume of distribution; CL, clearance; CL₂, intercompartment clearance.

Model evaluation.

A model diagnostics goodness-of-fit plot showed acceptable results for the final model. As presented in Fig. 3A to D, the population predicted concentrations (PRED) and individual prediction concentrations (IPRED) were evenly distributed around the diagonal line, implying no structural deviation in terms of visual biases. In the plots of conditional weighted residuals (CWRES) versus PRED and time after dose (TAD), most CWRES data were evenly distributed around the zero line and within the range of –2 and +2, indicating no significant systematic deviations in the model fit. Predicted-corrected visual predictive checks (VPCs) with the 90% confidence intervals (CIs) determined by using the final PopPK model graphed with the DV for linezolid are shown in Fig. 3E. On the basis of the predicted-corrected VPC plots, the final model adequately predicted the DV. Most of the observed 5th, 50th, and 95th quantiles were within the 90% CIs of the predicted corresponding quantiles, thus confirming the predictive performance of the developed model.

FIG 3.

Model evaluation for linezolid PopPK final model. (A) DV versus IPRED; (B) PRED (the solid line is diagonal). (C) CWRES versus TAD; (D) CWRES versus PRED (the blue line is trend line, and the red line is trend line of absolute CWRES). (E) Predicted-corrected VPC profile The red lines from bottom to top are the 5%, 50%, and 95% percentiles of DV, and the shaded area is the simulation-based 90% CIs for corresponding percentiles. DV, observed concentrations; PRED, population predicted concentrations; IPRED, individual predicted concentrations; CWRES, conditional weighted residuals; TAD, time after dose; VPC, visual predictive check; IVAR, independent variable. The circles represent DV.

The reliability and stability of this PopPK model was confirmed with a nonparametric bootstrap method. As shown in Table 2, the mean population parameter estimates obtained from the bootstrap procedure were generally comparable with those from the final model, indicating minimal bias. Furthermore, all the estimates of the structural parameters and random effects from the final model fell in the 95% CIs of the corresponding parameters obtained by the 1,000 bootstrap runs. This finding indicates that the final model was fairly robust and is therefore reliable and stable.

Model-based simulation and dosage optimization.

The PTA of different linezolid dosing regimens with an AUC/MIC of >80 was evaluated at MICs ranging from 0.125 mg/liter to 32 mg/liter by Monte Carlo simulation. The results are given in Fig. 4. The PTAs of the AUC/MIC target value of >80 at the steady state for the MICs of 1, 2, and 4 mg/liter were 96.9%, 63.6%, and 18.3% for the dose of 10 mg/kg every 8 h, respectively, and 100.0%, 98.3%, and 84.9% for the dose of 600 mg every 12 h, respectively. A higher dose of 15 mg/kg every 6 h reached the treatment target, resulting in the PTA of 94.6% for the MIC of 2 mg/liter. The AUC/MIC (assuming a MIC of 1 mg/liter) varied within the wide range of 40.85 to 461.82. The incidence of thrombocytopenia was significantly higher in patients with an AUC/MIC of >120 than in those with an AUC/MIC of 80 to 120 (53.49% versus 18.18%, P < 0.05). For children with normal (n = 27) and impaired (n = 36) liver function, the median AUC was 159.96 and 205.45 (P < 0.05), respectively. Fig. 5 illustrates the PTA as a function of different doses and AST levels. A dose of 10 mg/kg every 12 h was suitable for children aged <12 years, and 600 mg every 48 h was appropriate for those aged 12 to 18 years with an AST of >200 U/liter at a MIC of 1 mg/liter.

FIG 4.

PTA for MICs ranging from 0.125 mg/liter to 32 mg/liter as a function of different doses.

FIG 5.

PTA for MICs of 0.5, 1, 2, and 4 mg/liter as a function of different doses and AST levels ((A) 40 U/liter < AST ≤ 200 U/liter; (B) 200 U/liter < AST ≤ 400 U/liter; (C) AST > 400 U/liter).

DISCUSSION

To the best of our knowledge, this study is the first to evaluate the PopPK and PD of linezolid in children with serious infections admitted to the PICU. The work conducted in this distinctive population aimed to determine the linezolid PK properties, quantify the effects of demographic, clinical, and biological factors on linezolid disposition, and optimize the dosing regimens. A two-compartment model with first-order (linear) elimination was found to accurately describe the PopPK model. Weight and AST were the two significant covariates for clearance in the final model.

Several studies of linezolid PopPK indicated various results. Concentration data were described by one- or two-compartment models with parallel linear or nonlinear elimination at steady state following intravenous and oral administrations (9, 10). Abe et al. proposed a one-compartment model with linear elimination for data set analysis at the steady state following the intravenous and/or oral route (11). A two-compartment model with only linear elimination was also developed by using intravenous concentration data (12). However, no difference was observed between linear and nonlinear eliminations from the central compartment in two-compartment models using single-dose concentrations (13).

In this study, the two-compartment model with first-order (linear) elimination showed the best fit. Possible explanations lie in the following different conditions: the present data were obtained from children, but those in previous studies were all acquired from adults. The present data were obtained at steady state following the intravenous route, and the infusion interval was ≥1 h in each pediatric patient. Although previous studies suggested the one-compartment model rather than the two-compartment model for a short distribution phase after multiple dosing (14) and a long infusion interval (11), the reason was the PK difference between children and adults.

The PK of linezolid has been also described as a one-compartment model in pediatric patients (15). The ages of the study population in the two studies ranged from 0.03 to 11.9 years (15) and 0.1 to 15.3 years. Whether the individual patient was a critically ill child was also not considered. In the present work, additional samples were collected at different time points, except for trough concentrations, to accurately fit the PopPK of linezolid. The different time points of concentrations may contribute to this difference.

Earlier studies showed that linezolid clearance depends on renal function (10, 16–18). These reports were either PopPK analyses in adult patients or PK analyses in only five pediatric patients. Recent work reported a PopPK model of intravenous linezolid in 112 children (15) and revealed that weight and estimated glomerular filtration rate elucidated important portions of the variance. To date, the clearance pathways of linezolid have not been fully understood. Under steady-state conditions, approximately 30% of the drug dose appears in the urine as linezolid and 50% as metabolites. The changes of liver and renal function may affect linezolid PK.

In this study, the final model showed that weight and AST seriously influence the linezolid clearance in pediatric patients. Liver metabolism is a major eliminative route of linezolid (1); hence, hepatic function has a prominent effect on linezolid clearance and dosage regimen in pediatric patients. AST, a surrogate of hepatic function, independently influences linezolid clearance. Patients with an AST of >40 U/liter had a significantly higher linezolid AUC than those with an AST of ≤40 U/liter (205.45 versus 159.96, P < 0.05). Linezolid clearance was reduced in patients with hepatic dysfunction, thereby increasing their plasma drug concentration. Therefore, patients with impaired liver function required dose adaptation according to the different AST levels.

According to the CLSI and microbiology data provided in the labeling (27), the MIC₉₀ of linezolid for Staphylococci, Enterococci, and Streptococcus pneumoniae is 2, 2, and 0.5 mg/liter, respectively (19). This choice allows the risk of linezolid-related hematotoxicity to be contained (20). The MICs of 1 mg/liter were most frequently used and thus were selected in dosage optimization to cover pathogens in serious infections. The AUC/MIC ratio of >80 was the best threshold for linezolid efficacy and clinical outcome (3, 4). In the present work, a target AUC/MIC value of >80 was selected for dosage optimization. Modeling results showed the high variability of the AUC/MIC values in the study population with a range of more than 10-fold.

The PTA of the standard dosing regimen was above 90% for a MIC of ≤1 mg/liter, indicating that a successful clinical response was achieved in critically ill children. However, higher dosages may be needed, especially in the presence of pathogens with a MIC of >1 mg/liter. An optimal antimicrobial therapy should balance efficacy and toxicity (21). In this study, Monte Carlo simulations were conducted using the estimated parameters and the variability from the established PopPK model. The dosing regimen was optimized on the basis of achieving the target AUC/MIC value of >80. The results showed that the recommended age-differentiated standard dosing regimens of linezolid used in the study achieved the PK/PD target value. Most patients did not experience linezolid-related thrombocytopenia, and none needed linezolid withdrawal.

The risk frequency of thrombocytopenia was significantly higher in patients with a linezolid AUC/MIC of >120. The relationship between linezolid PK/PD (AUC/MIC) and thrombocytopenia, which is a major adverse reaction associated with linezolid treatment, was previously assessed. A higher PK/PD index allows for favorable outcomes for therapy, but a daily dose of linezolid of up to 60 mg/kg increases the risk of drug-related hematotoxicity (15). In such cases, clinical symptoms and hematological parameters should be monitored closely during treatment to avoid linezolid-related overexposure and toxicity.

In this study, 15 patients received continuous renal replacement therapy (CRRT) and 2 patients received extracorporeal membrane oxygenation (ECMO). These special situations were analyzed as covariates. The findings showed no significant influence on PK parameters. A definitive conclusion cannot be drawn on this aspect because whether the condition of CRRT or ECMO does not influence the PK of linezolid or sample size is limited remains unclear. Large clinical studies are clearly warranted to provide definitive evidence.

This study had some limitations. The PK model of linezolid was only internally validated. External validation was not performed, because of the limited number of patients. Ultimately, the optimal individual strategy based on modeling and simulation should be evaluated in clinical practice to confirm its clinical benefits (22). The model-based optimized dose regimens should be further evaluated in a large number of critically ill pediatric patients. In addition, the safety of simulated dosing regimens was not analyzed, thus warranting future studies.

Conclusion.

In this study, a PopPK model of linezolid was successfully developed for pediatric intensive care patients. Linezolid clearance was consistently correlated with body weight and inversely correlated with AST. The approved dosing regimens of linezolid were sufficient against sensitive pathogens but were inadequate against pathogens with a MIC of >1 mg/liter. An increased dose or alternative antimicrobial therapies may be required to cure infections with higher MICs in the specific population. Dose reduction is also necessary for children with impaired liver function. The evidence-based dosing regimen of linezolid in critically ill children was established based on PopPK/PD analysis.

MATERIALS AND METHODS

Study design.

The trial was a single-center, prospective, open-labeled PK study of linezolid performed in the PICU in Beijing Children’s Hospital. Pediatric patients aged 29 days to 18 years with suspected or confirmed MR-GPC infections receiving intravenous linezolid as part of their routine antimicrobial therapy were enrolled. This study was approved by the ethics committee of the hospital, and written informed consent was obtained from the guardians of each child prior to the work.

Dosing regimen, PK sampling, and data collection.

Linezolid (Pfizer Pharmaceuticals, USA) was intravenously infused by a syringe pump linked to a microbore valve, administered within 1 to 2 h at a dose of 10 mg/kg three times per day for patients aged <12 years and 600 mg twice daily for those aged 12 to 18 years. Four samples were collected at steady state from each patient half an hour before the infusion (trough concentration) and after the end of the infusion (peak concentration). Another two sample time points were constantly changed between peak and trough times (intermediate concentration). The times of infusion and sampling were accurately recorded. At each time point, a maximum of 2 ml of whole blood was collected in EDTA-anticoagulant-containing (purple-top) tubes. The samples were centrifuged at 3,000 rpm for 15 min at 4°C and stored at −70°C until analysis. Plasma concentrations were determined within 5 days after sampling. Clinical and demographic data, including age, sex, current weight, body mass index (BMI) calculated as weight (in kilograms) divided by height (in meters) squared, albumin (ALB), aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin (TB), direct bilirubin (DB), creatinine clearance (Ccr), pediatric critical illness score (PCIS), and receiving continuous renal replacement therapy (CRRT) or extracorporeal membrane oxygenation (ECMO) were extracted from the medical records of the pediatric patients in this study.

Linezolid assay.

The linezolid concentration was measured using a previously described method based on ultraperformance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) (23). The calibration curve was linear within the range of 0.078 mg/liter to 25.000 mg/liter. The lower limit of quantification (LLOQ) was identical to the lowest calibration levels. Concentrations higher than the quantitative curve were tested after dilution with blank plasma. The inter- and intraday coefficients of variation of controls ranged from 2.3% to 3.5%. The accuracies of controls were 99.8%, 99.2%, and 100.7%, and the matrix factors were 94.5%, 98.6%, and 93.5%.

Model development.

All plasma concentration-time data for linezolid were analyzed simultaneously with a population compartmental pharmacokinetic modeling approach using Phoenix NLME version 8.2 (Certara L.P., USA). First-order conditional estimation (FOCE) with interaction method was used to estimate pharmacokinetic parameters and their variability. The interindividual variability of PopPK parameters was estimated using the following exponential error model: Pi = P_TV · exp (ηi), where Pi is the individual pharmacokinetic parameter value for the ith subject, P_TV is the typical value for population, and η_i is the proportional difference between subjects and is supposed to be a normal distribution with a mean of 0 and a variance of ω².

Residual variability was evaluated using different error models, including additive (C_obs = C_pred + ε), multiplicative (Cobs = Cpred · [1 + ε]), exponential (Cobs = Cpred · [1 + ε1] + ε), and combinational models. Residual error was assumed to be normally distributed with a mean of 0 and a variance of σ². A stepwise forward-inclusion and backward-elimination procedure was followed for the covariate analysis. A likelihood ratio test was used to evaluate the statistically significant variables on the constructed model parameters. Covariates that showed a correlation with the individual PK parameters obtained from the basic model were successively introduced into the model. The potential effects of variables, including age, sex, weight, BMI, ALB, ALT, AST, TB, DB, Ccr, PCIS, and receiving CRRT or ECMO on PK parameters of V, V₂, CL, and CL₂ were examined.

Covariates were screened using a standard stepwise method. During the forward stepwise process, the covariates were added to the PopPK model and considered statistically significant when the OFV decreased (P < 0.05; χ² distribution; 1 degree of freedom; OFV reduction, >3.84). All significant covariates were synchronously incorporated into the basic model to create the full model. In the backward elimination step, each covariate was independently excluded from the full model on the basis of P < 0.01 (OFV increase, >6.63; 1 degree of freedom).

Model qualification.

The final model was validated using goodness-of-fit plots and a bootstrap approach. First, the predictive accuracy of the final pharmacokinetic model was examined using VPCs (24), which were performed by simulating the plasma concentration-time profiles for linezolid using Phoenix NLME software. A total of 1,000 simulations were conducted to create plasma concentration-time profiles of linezolid by using the data from all included patients to build the PopPK model. Curves for the 5th, 50th, and 95th percentiles of simulated drug concentrations were graphed with the observed concentrations (dependent variable [DV]). Residual plots using conditional weighted residuals (CWRES) versus time and population prediction concentrations (PRED) were checked (25). The reliability and stability of the final PopPK model was assessed by a nonparametric bootstrap method using resampling and replacement methods. Resampling was repeated 1,000 times, and the parameter values predicted from the bootstrap approach were further contrasted with those derived from the original data set and individually fitted to the final PopPK model. All model parameters were estimated, and their bootstrap 95% CIs were calculated.

Simulation and dosing regimen optimization.

The final PopPK model was used to simulate linezolid concentrations at different time points. Monte Carlo simulations were performed using Phoenix NLME to optimize the dose strategy that can achieve the PTA at an AUC/MIC value of >80 for different dose regimens (the dose in mg/kg in children <12 years and the dose in mg in older children) for MICs of 0.125 to 32 mg/liter, and a PTA of ≥90% was considered optimal against that bacterial population (26). A linezolid MIC value of 1 mg/liter was selected for dosage optimization simulation to cover pathogens. Platelet counts of ≥100 × 10³/μl were considered to ensure comparable safety conditions. Various dosing regimens were simulated in hypothetic clinical trials for critically ill children with normal liver function (AST, ≤40 U/liter) and impaired liver function (AST, >40 U/liter). Simulations were performed 1,000 times using the final PopPK model, and the AUC at steady state was calculated using noncompartmental analysis method based on the concentration data obtained from Monte Carlo simulation for every simulated patient. The percentage of patients achieving an AUC/MIC target ratio of >80 was subsequently calculated to optimize the appropriate dose regimen in each subgroup.