Population Pharmacokinetics and Dosing Optimization of Imipenem in Children with Hematological Malignancies

Imipenem is widely used for the treatment of children with serious infections. Currently, studies on the pharmacokinetics of imipenem in children with hematological malignancies are lacking.

ABSTRACT

Imipenem is widely used for the treatment of children with serious infections. Currently, studies on the pharmacokinetics of imipenem in children with hematological malignancies are lacking. Given the significant impact of disease on pharmacokinetics and increased resistance, we aimed to conduct a population pharmacokinetic study of imipenem and optimize the dosage regimens for this vulnerable population. After children were treated with imipenem-cilastatin (IMP-CS), blood samples were collected from the children and the concentrations of imipenem were quantified using high-performance liquid chromatography with UV detection. Then, a population-level pharmacokinetic analysis was conducted using NONMEM software. Data were collected from 56 children (age range, 2.03 to 11.82 years) with hematological malignancies to conduct a population pharmacokinetic analysis. In this study, a two-compartment model that followed first-order elimination was found to be the most suitable. The parameters of current weight, age, and creatinine elimination rate were significant covariates that influenced imipenem pharmacokinetics. As a result, 41.4%, 56.1%, and 67.1% of the children reached the pharmacodynamic target (the percentage of the time during the total dosing interval that the free drug concentration remains above the MIC of 70%) against sensitive pathogens with an MIC of 0.5 mg/liter with imipenem at 15, 20, and 25 mg/kg of body weight every 6 h (q6h), respectively. However, only 11.1% of the children achieved the pharmacodynamic target against Pseudomonas aeruginosa isolates with an MIC of 2 mg/liter at a dose of 25 mg/kg q6h. The population pharmacokinetics of imipenem were assessed in children. The current dosage regimens of imipenem result in underdosing against resistant pathogens, including Pseudomonas aeruginosa and Acinetobacter baumannii. However, for sensitive pathogens, imipenem has an acceptable pharmacodynamic target rate at a dosage of 25 mg/kg q6h. (The study discussed in this paper has been registered at ClinicalTrials.gov under identifier NCT03113344.)

TEXT

Imipenem is used for the treatment of patients with polymicrobial infections of the respiratory system, urinary system, intra-abdominal region, reproductive system, bones and joints, and skin and with bacterial septicemia (1). It is a highly efficient, broad-spectrum, and FDA-approved drug that is used to treat children and infants <3 months of age suffering from serious infections caused by Gram-positive and Gram-negative bacteria (2, 3).

Imipenem is a lipophilic drug that can rapidly penetrate various body cavities, including the respiratory tract, tissue in the abdominal cavity, and soft tissue (4). As a combination antibiotic product, each agent in the IMP-CS combination has a short plasma half-life of about 1 h and low protein binding (20% and 40% for imipenem and cilastatin, respectively) (5). In patients with normal renal function, renal elimination accounts for 70% of imipenem removal (5). In adults, the percentage of the time during the total dosing interval that the free drug concentration remains above the MIC (%fT_>MIC) is an important pharmacokinetic (PK) index for the estimation of therapeutic efficacy (6). Despite having extensive applications in clinical treatment, pharmacokinetic data on imipenem are scarce and remote for children, especially for special pediatric patients, such as children with hematological malignancies, for whom the pharmacokinetics have been shown to differ significantly from those for other patients (7, 8). The half-life of imipenem in patients with renal failure is two to three times that in individuals with normal renal function (9). Insufficient pharmacokinetic data are likely to increase the risk of antibiotic treatment failure or the probability of drug resistance in children (10). Therefore, the purpose of this study was to evaluate the pharmacokinetics of imipenem in children with hematological malignancies through a population method and provide optimized dosage regimens of imipenem for use in children.

RESULTS

Study population.

This study initially included 56 patients admitted to the hospital between 2016 and 2017, and informed consent was obtained at the same time. Of the total of 56 patients, 12 had two hospitalizations in this study. All patients fully met the inclusion and exclusion criteria. All of them received the imipenem dose regimen of 15 to 25 mg/kg of body weight every 6 h (q6h). No patients withdrew from this trial due to adverse events. The mean current weight (CW) of the patients in the study was 18.65 kg (standard deviation [SD], 6.90 kg). The characteristics of the patients included are presented in Table 1.

TABLE 1.

Baseline characteristics of the children included in the study^c

Characteristic	Mean (SD) value	Median (range) value
Age (yr)	4.86 (2.33)	4.69 (2.03–11.82)
CWa (kg)	18.65 (6.90)	18.00 (10.00–44.00)
CLCRb (ml/min)	223.2 (79.5)	241.0 (95.1–668.6)
Imipenem dose
mg/dose	348.3 (135.1)	300.0 (210.0–880.0)
mg/kg/dose	18.7 (2.4)	20.0 (13.2–25.0)

^aCW, current weight.

^bCL_CR, creatinine clearance rate. CL_CR was calculated by the Schwartz formula.

^cData are for 56 patients (30 males and 26 females).

Model building.

There were 136 imipenem concentrations used for population modeling. The concentration of imipenem in the samples ranged from 0.5 to 88.42 μg/ml. The concentration-time curve of imipenem is shown in Fig. 1.

FIG 1.

Imipenem concentration versus time.

It was found that a two-compartment model following first-order elimination was suitable for the data set. The two-compartment model had lower objective function values (OFV) and residual variability than a one-compartment model. The parameters central volume of distribution (V₁), peripheral volume of distribution (V₂), intercompartment clearance (Q), and clearance (CL) of imipenem were simulated in the model. The exponential model could best describe the interindividual variability, which was further estimated for V₁ and CL. The combined additive and proportional model was used to describe residual variability.

Covariate analysis.

A significant drop of 30.8 points in the OFV was obtained by incorporating CW a priori into the basic model (allometric coefficients, 0.75 for CL and Q, 1 for V₁ and V₂) in the allometric size approach. Then, the OFV decreased by 10.28 and 15.04 units by implementing age and CL_CR on clearance, respectively. The CL_CR was calculated by the Schwartz formula. There were no other tested covariates that caused remarkable optimization. The process of covariate analysis is shown in Table S1 in the supplemental material. Therefore, the covariates CW, age, and CL_CR were included in the final model.

At steady state (sum of V₁ and V₂), CL normalized by weight and the volume distribution were simulated to have median values of 0.47 liters/h/kg (range, 0.26 to 0.95 liters/h/kg) and 0.76 liters/kg (range, 0.73 to 0.78 liters/kg), respectively. The relationship between imipenem CL and CW, age, and CL_CR is shown in Fig. 2A to C.

FIG 2.

Relationship between imipenem CL and CW (A), age (B), and CL_CR (C).

Model evaluation.

For the final model of imipenem pharmacokinetics, an acceptable goodness of fit was obtained through model diagnostics. The predictions in Fig. 3A and B were unbiased. No trends in the scatter diagrams of conditional weighted residuals (CWRES) versus time and population prediction (PRED) in Fig. 3C and D were observed. In addition, the median parameter estimates derived from the bootstrap procedure closely aligned with the corresponding values from the final model and could be used to redetermine the population pharmacokinetic parameter estimates in a stable manner (Table 2). At the same time, the results of the normalized prediction distribution errors (NPDEs) are shown in Fig. 3E and F. The distribution and histogram of NPDE satisfied the theoretical N(0,1) distribution and density well, which proved that the model was highly suitable for individual data. In this study, the mean and variance of NPDE were 0.0534 (P = 0.522 in the Wilcoxon signed-rank test) and 1.1 (0.395 in the Fisher variance test), respectively. The result of the prediction-corrected visual predictive check is shown in Fig. 3G. The prediction-corrected concentrations obtained suitably matched the simulated concentrations, confirming the predictive effect of the final model.

FIG 3.

Model evaluation for imipenem. (A) PRED versus DV. (B) IPRED versus DV. (C) CWRES versus time. (D) CWRES versus PRED. (E) QQ plot of the distribution of NPDE versus the theoretical N(0,1) distribution. (F) Histogram of the distribution of NPDE, with the density of the standard Gaussian distribution overlaid. (G) Prediction-corrected visual predictive check. The circles represent the prediction-corrected observed concentrations. The solid line represents the median prediction-corrected observed concentrations. The observed 5% and 95% percentiles are presented with dashed lines.

TABLE 2.

Population pharmacokinetic parameters for imipenem and bootstrap results^a

Parameter	Full data set		Bootstrap analysis
	Final estimate	RSE (%)	Median	5th percentile-95th percentile
V1 (liters), where V1 = θ1 × (CW/18), θ1	7.20	6.40	7.19	6.28–8.06
V2 (liters), where V2 = θ2 × (CW/18), θ2	6.51	22.9	8.64	1.02–345.9
Q (liters/h), where Q = θ3 × (CW/18)0.75, θ3	0.996	28.2	1.21	0.147–5.57
CL (liters/h), where CL = θ4 × (CW/18)0.75 × Fage × FCLCR′, θ4	8.60	5.50	8.41	6.37–9.36
Fage = (age/4.69)θ5, θ5	0.265	36.8	0.293	0.101–0.668
FCLCR = (CLCR/214)θ6, θ6	0.509	35.4	0.479	0.224–0.803
Interindividual variability (%)
CL	18.8	52.7	20.9	0.894–40.2
V1	9.20	134.8	7.65	0.775–20.0
Residual variability (%)
ERR(1)	39.5	17.9	37.5	31.2–44.6
ERR(2)	20.5	47.2	18.0	0.707–27.7

^aIn the first 6 rows of the table, the values in columns 2 to 5 refer to θ1 to θ6. V₁, central volume of distribution; V₂, peripheral volume of distribution; Q, intercompartmental clearance; CL, clearance; CW, current weight (in kilograms); CL_CR, creatinine clearance rate (CL_CR was calculated by the Schwartz formula); F_age, scaling factor applied for patients coadministered with age; F_CLCR′, scaling factor applied for patients coadministered with CL_CR; RSE, relative standard error; ERR, error.

Dosing regimen optimization.

The probability of target attainment (70% fT_>MIC) at MICs of 0.125, 0.5, 2, and 4 mg/liter are shown for different imipenem doses (15, 20, 25 mg/kg/per dose) in Table 3. Common pathogens, including Escherichia coli, Klebsiella pneumoniae, and Staphylococcus aureus (methicillin sensitive), with MICs of ≤0.5 mg/liter are sensitive to imipenem (11, 12). According to EUCAST, the PK/pharmacodynamic (PD) MIC breakpoint of imipenem against sensitive pathogens is 2 mg/liter (13). The probability of target attainment (70% fT_>MIC) for standard MIC susceptibility breakpoints at 0.5 mg/liter and 2 mg/liter is shown in Fig. 4. When using the dosing regimens prescribed in this study, 41.4%, 56.1%, and 67.1% of the children reached the pharmacodynamic target (70% fT_>MIC) for an MIC of 0.5 mg/liter at 15, 20, and 25 mg/kg imipenem q6h, respectively. For an MIC of 2 mg/liter at 25 mg/kg imipenem q6h, only 11.1% of the children reached the pharmacodynamic target (70% fT_>MIC).

TABLE 3.

Probability of target attainment at different MICs^a

MIC (mg/liter)	Probability of target attainment (%) at the following imipenem dose (mg/kg/per dose):
	15	20	25
0.125	84.2	90.1	93.9
0.500	41.4	56.1	67.1
2.00	1.64	5.41	11.1
4.00	0.0893	0.411	1.23

^aThe probability of target attainment was an fT_>MIC of 70%.

FIG 4.

Probability of target attainment (70% fT_>MIC) for imipenem at different dose regimens with MICs of 0.5 mg/liter and 2 mg/liter.

DISCUSSION

The present work is the first imipenem population pharmacokinetic study conducted in children with hematological malignancies. Along with first-order elimination, a two-compartment model was optimal for the modeling of our pharmacokinetic data. In this study, the value of the estimated CL median was 0.47 liter/h/kg, which seems to be similar to values that have been previously published for children but which are significantly lower than the values for adults (8, 14, 15). The difference can be explained well by renal maturation. As a combination with cilastatin at a fixed ratio of 1:1, imipenem is extensively metabolized by dehydropeptidase I (DHP-I), resulting in the excretion of about 70% of the unchanged drug into the urine in the renal system (16, 17). In an adult population pharmacokinetic study of imipenem, analysis of the data found that clearance increased along with CL_CR (18). In contrast, none of the tested covariates, including CL_CR, were found to have any identifiable influence on the pharmacokinetic parameters of imipenem in a study conducted in Thailand (19).

In the 1980s and 1990s, several studies paid attention to imipenem pharmacokinetics in newborns, infants, and children (7, 8, 20). The data for the pharmacokinetic parameters revealed differences from the data for adult subjects. However, the results of these existing pediatric studies were limited by the population pharmacokinetic simulation technology and hardly provided valid evidence for current dose regimens. In our study, CW, age, and CL_CR were identified to be important factors influencing the population pharmacokinetics of imipenem in children. The impact of CL_CR has been reported for other drugs, such as ciprofloxacin and vancomycin, which are primarily eliminated through the renal pathway (21–23). None of the other covariates tested affected the pharmacokinetic parameters in the processes of both forward and backward selection.

The therapeutic window of opportunity for successful treatment according to clinical experience begins with infection management. The method used to forecast the probability of reaching the target drug exposure through a pharmacokinetic model, as well as microbiological variability, is increasingly important. Imipenem is an important member of the carbapenems and is a time-dependent antibiotic. As a parameter important for pharmacokinetics (PK)/pharmacodynamics (PD), the %fT_>MIC is correlated with therapeutic efficacy (6). The 40% fT_>MIC PK/PD target is commonly used and is well characterized in imipenem studies (15, 18, 19). Although there were few imipenem pharmacodynamic data for children prior to our study, the 70% fT_>MIC is a more conservative endpoint.

In several Chinese studies, imipenem has shown good antimicrobial activity against sensitive pathogens, such as E. coli and K. pneumoniae, which have low rates of drug resistance. Surprisingly, the rates of resistance to imipenem by Pseudomonas aeruginosa and Acinetobacter baumannii appear to have shown an upward trend in recent years and have rapidly increased to more than 80% and nearly 50%, respectively (24–26).

Based on our simulation results, the 25-mg/kg q6h dosing regimen of imipenem provided a 67.1% probability of target attainment at a 70% fT_>MIC against sensitive pathogens (MICs ≤ 0.5 mg/liter). However, at a 70% fT_>MIC, a probability of target attainment of only 11.1% against P. aeruginosa (MIC = 2 mg/liter) was obtained with 25-mg/kg q6h imipenem. In addition, imipenem was inactive against resistant pathogens, such as A. baumannii (MICs ≥ 16 mg/liter), in this study. Therefore, the current 25-mg/kg q6h dosing regimen for imipenem indicated on drug labels cannot easily meet the needs for the treatment of infections caused by pathogens other than sensitive pathogens (MICs ≤ 0.5 mg/liter).

Our study has some limitations. Clinical efficacy data were not collected in the present study due to the limited number of patients and the complex antimicrobial therapies. The model-based optimized dose should therefore be evaluated further.

Conclusions.

A population pharmacokinetic model of imipenem was developed for children with hematological malignancies. Current weight, age, and renal clearance had significant impacts on the pharmacokinetics of imipenem. The current dosage regimens of imipenem were able to achieve the pharmacodynamic breakpoint for sensitive pathogens but achieved underdosing against resistant pathogens, especially P. aeruginosa and A. baumannii. An increased dose or alternative antimicrobial therapies may be required to cure P. aeruginosa and A. baumannii infections in children.

MATERIALS AND METHODS

Study design.

This pharmacokinetic study of imipenem was a prospective and open-label trial that was conducted at the Children’s Hospital of Hebei Province, Shijiazhuang, China. Inclusion and exclusion criteria are presented in Fig. 5. This study was designed in accordance with Chinese legal requirements and the Declaration of Helsinki and was approved by the institutional ethics board of the Children’s Hospital of Hebei Province.

FIG 5.

Inclusion and exclusion criteria.

Dosing regimen and pharmacokinetic sampling.

IMP-CS (Haihui; Haizheng Pfizer Pharmaceuticals Corporation, Hangzhou, China) was administered at 15 to 25 mg/kg intravenously over 30 to 60 min four times daily (q6h). The dosage recommendation represents the quantity of imipenem to be administered. The total blood sampling frequency for every participant was limited to 2 times at steady state. All patients included were sampled according to two predefined pharmacokinetic sampling schedules: (i) 3 to 5 min after the end of infusion, as well as 2 to 4 h after the start of infusion, and (ii) 0.5 to 1 h after the end of infusion, as well as 4 to 6 h after the start of infusion. Imipenem pharmacokinetics were assessed on the first or second day of treatment. The precise time of infusion and blood sample collection was recorded. No less than 0.5 ml of blood was sampled each time. The validated sampling information for the samples was necessary to be included in this study. Blood samples were refrigerated and centrifuged (2,500 × g at 4°C for 10 min), and the plasma sample obtained was then frozen at −70°C. Finally, samples were transported in dry ice to the Clinical Pharmacy Department of Shandong Provincial Qianfoshan Hospital and then rapidly stored at −80°C in a freezer prior to analysis.

Method of imipenem analysis.

In brief, using metronidazole as an internal standard, the concentration of imipenem was measured using high-performance liquid chromatography (HPLC) with UV detection.

Analytical procedures were performed on an LC-2030C HPLC system, and the compounds were separated on an InertSustain C₁₈ column (4.6 by 250 mm; particle size, 5 μm), which was purchased from Shimadzu (Kyoto, Japan), connected to a UV detector at 300 nm for imipenem. A mobile phase consisted of MOPS (morpholinepropanesulfonic acid) buffer (0.004 M, pH 7.00, containing 0.2% sodium hexane sulfonate) (mobile phase A) and acetonitrile (mobile phase B) in a gradient mode of elution at a flow rate of 0.7 ml/min at 40°C. The calibration curve obtained ranged from 0.5 to 50 μg/ml. The inter- and intraday coefficients of variation (CVs) for the controls were under 9%. In addition, the lower limit of quantification (LLOQ) was 0.5 μg/ml for imipenem.

Population pharmacokinetic modeling of imipenem.

The pharmacokinetic analysis process was completed using the nonlinear mixed effects modeling (NONMEM) program (v7.2; Icon Development Solutions, USA). The pharmacokinetic parameters and corresponding variability were estimated through the method of first-order conditional estimation (FOCE).

An exponential model was selected to evaluate the interindividual variability of the pharmacokinetic parameters and is indicated as follows: θ_i = θ_mean · e^ηi, where θ_i stands for the value of the parameter that belongs to the ith subject, θ_mean is the typical value of the corresponding population parameter, and ηi is the subject’s variability with a mean value between zero and variance ω2 that is supposed to agree with a normal distribution.

Covariate analysis was performed after a forward and a backward selection course. For each variable, the effect on the model parameters was calculated using the likelihood ratio test. The influences of current weight (CW), age, and CL_CR were explored as potential variables. First, a covariate was included in the covariate model by establishing the point at which the objective function value (OFV) of the basic model significantly decreased (P < 0.05, χ² distribution with 1 degree of freedom) by more than 3.84 and variability in the pharmacokinetic parameter was found to be reduced. The covariates with a significant reduction were contemporaneously incorporated into a full model. Next, each covariate was independently eliminated from the full model. If the covariate was identified to be significantly associated with the pharmacokinetic parameter and if the OFV increased by more than 6.635 (P < 0.01, χ² distribution), it was retained in the final model.

Graphical and statistical criteria were applied to validate the model. Goodness-of-fit plots consisted of the observed value (DV) versus the population prediction (PRED), DV versus the individual prediction (IPRED), conditional weighted residuals (CWRES) versus time, and CWRES versus PRED, which were originally used for diagnostic purposes (27). In the final model, the stability and properties were also evaluated through resampling and replacement by nonparametric bootstrap analysis. Resampling was repeated up to 1,000 times, and the parameter values estimated after bootstrap analysis were compared with the values measured from the initial data set. Normalized prediction distribution errors (NPDE) were used to statistically and graphically assess the final model (28, 29). One thousand data sets were simulated using the parameters of the final population model. The results of NPDE were summarized by default in graphical form using the NPDE R package (v1.2) as a (i) quantile-quantile (QQ) plot of the NPDE and (ii) a histogram of the NPDE. In addition, NPDE was expected to comply with an N(0,1) distribution (30).

Dosing regimen optimization.

In order to gain the greatest bactericidal effect of carbapenems, a target attainment probability of 70% was required for at least 70% of the dosing interval (31). The population pharmacokinetic parameters which were estimated by the final model were used in a Monte Carlo simulation (32, 33). The pediatric dose of imipenem was simulated on the basis of the dose (in milligrams per kilogram) according to age. The dosing regimen (15 to 25 mg/kg imipenem q6h) applied for the original data set was simulated. One thousand simulations of the original data set were performed, and fT_>MIC was calculated for every simulated patient. For different dosage regimens, the probability of target attainment was calculated to optimize antimicrobial therapy.