Population pharmacokinetics and dosing optimization of teicoplanin in children with malignant haematological disease

Abstract

Aim

Children with haematological malignancy represent an identified subgroup of the paediatric population with specific pharmacokinetic parameters. In these patients, inadequate empirical antibacterial therapy may result in infection-related morbidity and increased mortality, making optimization of the dosing regimen essential. As paediatric data are limited, our aim was to evaluate the population pharmacokinetics of teicoplanin in order to define the appropriate dosing regimen in this high risk population.

Methods

The current dose of teicoplanin was evaluated in children with haematological malignancy. Population pharmacokinetics of teicoplanin were analyzed using nonmem software. The dosing regimen was optimized based on the final model.

Results

Eighty-five children (age range 0.5 to 16.9 years) were included. Therapeutic drug monitoring and opportunistic samples (n = 143) were available for analysis. With the current recommended dose of 10 mg kg^–1 day^–1, 41 children (48%) had sub-therapeutic steady-state trough concentrations (C_ss,min<10 mg l^–1). A two compartment pharmacokinetic model with first order elimination was developed. Systematic covariate analysis identified that bodyweight (size) and creatinine clearance significantly influenced teicoplanin clearance. The model was validated internally. Its predictive performance was further confirmed in an external validation. In order to reach the target AUC of 750 mg l^–1 h 18 mg kg^–1 was required for infants, 14 mg kg^–1 for children and 12 mg kg^–1 for adolescents. A patient-tailored dose regimen was further developed and reduced variability in AUC and C_ss,min values compared with the mg kg^–1 basis dose, making the modelling approach an important tool for dosing individualization.

Conclusions

This first population pharmacokinetic study of teicoplanin in children with haematological malignancy provided evidence-based support to individualize teicoplanin therapy in this vulnerable population.

What is Already Known about this Subject

• The pharmacokinetics of teicoplanin show large inter-individual variability, making therapeutic drug monitoring useful to optimize an individual dose, especially in children.

• Children with malignant haematological disease are a special group of patients with proper pharmacokinetic parameters, different from other paediatric patients. The pharmacokinetic data of teicoplanin are missing in this vulnerable population, thereby leading to the empirical therapy in clinical practice.

What this Study Adds

• With the current recommended dose of 10 mg kg⁻¹ day⁻¹, 48% of patients had sub-therapeutic steady-state trough concentrations (C_ss,min<10 mg l⁻¹). The dose needs to be increased.

• A population pharmacokinetic analysis has been conducted in 85 children. Body weight and creatinine clearance have been identified as significant covariates influencing teicoplanin clearance.

• A patient-tailored dose regimen was developed in children with malignant haematological disease, making the modelling approach an important tool for dosing individualization.

Introduction

The first line treatment for invasive methicillin-resistant Staphylococcus aureus (MRSA) infections is a glycopeptide antibiotic, either vancomycin (a glycopeptide) or teicoplanin (a lipoglycopeptide), acting by inhibition of peptidoglycan synthesis in the bacterial cell wall 1. Teicoplanin is often included in the initial empirical antibiotic therapy for febrile, neutropenic patients with haematological malignancy (HM), is not inferior to vancomycin with regard to efficacy and is associated with a lower adverse event rate than vancomycin 2,3.

Teicoplanin is not absorbed orally and is thus administered by the parenteral route (intravenously or intramuscularly). It is mainly bound to albumin (90%) and eliminated predominantly by the kidney 4. Some pharmacokinetic studies of teicoplanin have been conducted in children 5–7, however, only in children without HM. As demonstrated in our previous pharmacokinetic study of vancomycin, eight children with HM represent an identified sub-population of patients with proper pharmacokinetic parameters, different from other paediatric patients. In these patients, an inadequate drug dose regimen may result in infection-related morbidity and increased mortality, making optimization of the dosing regimen essential 8,9. Thus, the objectives of the present work were to evaluate the population pharmacokinetics of teicoplanin in children with HM using a modelling and simulation approach and to propose dosage optimization based on a pre-defined pharmacokinetic–pharmacodynamic breakpoint.

Methods

Study population

Children with HM receiving teicoplanin for suspected infection were studied in the Department of Paediatric Haematolo-Oncology at Robert Debré Hospital between 2012 and 2013. A two-step inclusion process was conducted. In the first step the population pharmacokinetic model was built using the pharmacokinetic data from 85 children and in the second step, an independent group of 15 children with similar characteristics and clinical condition were included for an external validation of the model. The following data were collected prospectively by a trained research assistant: age, weight, serum creatinine concentration, teicoplanin administration (dose and injection times), type of haematological disease and possibly bone marrow transplantation. All patients with incomplete dosing information were excluded. This study was designed in accordance with legal requirements and the Declaration of Helsinki, registered at the Commission Nationale Informatique et Liberté (CNIL) and approved by the local research ethics committee (Comité de l'Evaluation de l'Ethique des Projets de Recherche Biomédicale [CEERB], Robert Debré Hospital, Paris, France). All parents or guardians have been informed the study. According to the French law, no consent is required for routine samples.

Dosing regimen and sampling

Teicoplanin (Targocid, Sanofi-Aventis) was administered as an intravenous injection over 3–5 min. The initial dosing regimen was 10 mg kg^–1 every 12 h for three loading doses followed by maintenance dose of 10 mg kg^–1 once daily. Monitoring of teicoplanin concentrations was performed in order to maintain a steady-state trough concentration (C_ss,min) ≥ 10 mg l^–1.

Two types of serum samples were available, therapeutic drug monitoring (TDM) and opportunistic (also called scavenged) samples. The trough concentrations at steady-state (C_ss,min) were taken for TDM. During the whole teicoplanin treatment period, opportunistic samples were collected from blood remaining after routine biochemical tests. Tubes with specific research labels were used, allowing the laboratory staff to identify the opportunistic samples and store them at –70 °C after routine testing. The standard operation procedure to handle the opportunistic samples is the same as the pharmacokinetic samples in a traditional pharmacokinetic study. Precise sampling time and drug administration history (including dosing and infusion time) were recorded prospectively by the clinical team using dedicated documentation and transcribed later to the case report form. Samples without precise sampling time were excluded.

Assay of serum teicoplanin and creatinine

The serum teicoplanin concentrations were determined by the quantitative microsphere system (QMS) using the CDX automate (Thermo Fisher Scientific, Germany). The calibration curve ranged from 0 to 100 mg l^–1. The accuracy and coefficients of variation (CVs) of intra-laboratory controls (10, 35 and 75 mg l^–1) were 2.8% to 2.9% and 3.6%, respectively. The lower limit of quantification (associated CV) was 3 mg l^–1 (CV of <5%). Serum creatinine concentrations were measured by an enzymatic method using an Advia 1800 chemistry system (Siemens Medical Solutions Diagnostics, Puteaux, France).

Pharmacokinetic modelling

Pharmacokinetic analysis was carried out using the non-linear mixed effects modelling program nonmem, version 7.2.0 (Icon Development Solutions, Ellicott City, MD, USA). The first order conditional estimation (FOCE) method with the interaction option was used to estimate the pharmacokinetic parameters and their variability.

Structure model

One or two compartment open models with first order elimination were compared. The basic model was evaluated through visual inspection of routine diagnostic plots. Inter-individual variability of the pharmacokinetic parameters was estimated using an exponential model and was expressed as follows:

1

where θ_i represents the parameter value of the ith subject, θ_TV is the typical value of the parameter in the population and ηi is the variability between subjects, which is assumed to follow a normal distribution with a mean of zero and variance of 1.

A residual variability (additive, proportional, exponential, or mixed) model was selected according to improvement of the objective function value (OFV) and visual inspection of routine diagnostic plots.

Covariate analysis

Covariate analysis followed a forward-backward selection process and biological plausibility. The likelihood ratio test was used to test the effect of each variable on model parameters. The effects of age, weight, serum creatinine concentration, creatinine clearance (Schwartz formula) and type of HM (leukaemia or lymphoma) were investigated as potential variables on pharmacokinetic parameters. During forward selection, a covariate was selected if a significant (P < 0.05, χ² distribution with 1 degree of freedom) decrease (reduction of >3.84) in the OFV from the basic model was obtained. At the end, all the significant covariates were added simultaneously into a ‘full’ model. The importance of each covariate was reassessed by backward selection, and a covariate was independently removed from the full model if the increase in the OFV was less than 6.64 (P < 0.01, χ² distribution). The resulting model was considered the ‘final’ population pharmacokinetic model.

Model validation

Model validation was based on graphical and statistical criteria. Goodness-of-fit plots, including observed (DV) vs. individual prediction (IPRED), DV vs. population prediction (PRED), conditional weighted residuals (CWRES) vs. time and CWRES vs. PRED, were used initially for diagnostic purposes 10. The stability and performance of the final model were also assessed by means of a non-parametric bootstrap with resampling and replacement. Resampling was repeated 500 times, and the values of estimated parameters from the bootstrap procedure were compared with those estimated from the original data set. The entire procedure was performed in an automated fashion, using the PsN module 11. The final model was also evaluated graphically and statistically by normalized prediction distribution errors (NPDE) and prediction-corrected visual predictive check (pcVPC) 12,13. One thousand datasets were simulated using the final population model parameters. NPDE results were summarized graphically by default as provided by the NPDE R package (v1.2) 14: (i) QQ-plot of the NPDE and (ii) histogram of the NPDE. The NPDE is expected to follow the N (0, 1) distribution. For pcVPC, observed and simulated dependent variables were normalized based on the typical population prediction for the median independent variable in the bin. 95% confidence intervals for the median, 5^th and 95^th percentiles of the prediction-corrected simulated concentrations were calculated, plotted against time and compared with the prediction-corrected observed concentrations.

Given that the objective of the analysis was to use the final model for prediction purposes, an external evaluation was performed to evaluate the predictive performance of the developed model in an independent group of children with malignant haematological disease. The individual concentrations were predicted by Bayesian estimation (MAXEVAL = 0 in the estimation step, where MAXEVAL is the maximum number of model evaluations that can be used) with nonmem using the population pharmacokinetic parameters. The predictive performance was evaluated by calculating the prediction error (PE) and absolute prediction error (APE) using the following equations (ABS is the absolute function):

2

3

Dosing optimization based on a pharmacokinetic model

Monte Carlo simulations were performed using the parameter estimates obtained from the final model in order to define the optimal dosing regimen able to attain the target AUC of 750 mg l^–1 h in 50% of patients, under the assumption of a comparable safety profile.

Traditional paediatric dose (mg kg^–1 basis) simulation approach

In the traditional approach, the paediatric dose of teicoplanin was simulated on a mg kg^–1 basis according to different age groups. Thus, various mg kg^–1 dosing regimens (10, 11, 12, 13, 14, 15, 16, 17 and 18 mg kg^–1) were simulated in each paediatric group: infants (1 month to 2 years), children (2 to 12 years) and adolescents (12 to 18 years). One hundred simulations were performed using the original data set, and AUC(0,24 h) and C_ss,min values were calculated for each patient simulation. The target attainment rate was then calculated for each dosing regimen to define the optimal dose regimen in each paediatric group.

Patient-tailored dose

A patient-tailored dose was assessed to evaluate the advantage of personalized therapy. In this simulation scenario, the individual dose was calculated based on population pharmacokinetic parameters and covariates in each patient as follows:

4

where CL_i is calculated using the equation developed from the model, and i stands for individual. The simulation process was similar to that described above for dose simulation on the mg kg^–1 basis. At the end, we compared the variability of AUC(0,24 h) and C_ss,min values between the mg kg^–1 basis and patient-tailored dose.

Results

Serum concentrations of teicoplanin were measured in 85 children (53 boys) aged 0.5 to 16.9 years, weighing 7.7 to 90.6 kg. A summary of patient demographic and clinical characteristics is presented in Table1.

Table 1.

Baseline characteristics of 85 children

	Number	Mean	SD	Median	Range
Patients	85
Gender	32 F/53 M
Bone marrow transplantation	34
Samples	143
Teicoplanin dose (mg)		293	144	270	90–600
Teicoplanin dose (mg kg–1)		9.4	1.4	9.6	4.9–13.2
Teicoplanin concentration (mg l–1)		13.9	8.3	11.8	<LLOQ–49.6
Weight (kg)		32.3	17.8	27.1	7.7–90.6
Age (years)		8.4	4.6	8.1	0.5 – 16.9
Infants (1 month to 2 years)	10
Children (2 to 12 years)	49
Adolescents (12 to 18 years)	26
Serum creatinine (µmol l–1)		41	23	33	12–121
Creatinine clearance (ml min–1)*		191.2	76.2	178.9	48.6–464.1
Haematology disease
Acute lymphoblastic leukemia	41
Acute myeloblastic leukemia	27
Biphenotypic acute leukemia	4
Juvenile myelomonocytic leukemia	3
Lymphoma	5
Other	5

LLOQ, lower limit of quantification;

^*creatinine clearance was calculated by the Schwartz formula.

A total of 143 teicoplanin concentrations were either drawn for therapeutic drug monitoring (TDM, n = 123) or were opportunistic samples (n = 20). The concentrations ranged from <LLOQ to 35.1 mg l^–1. The numbers of patients in the concentration ranges for the first TDM sample (<10, 10 to 20, 20 to 30 and >30 mg l^–1) were 41 (48%), 32 (38%), 8 (9%) and 4 (5%), respectively.

Population pharmacokinetic modelling

A two compartment model with first order elimination fitted the data. The OFV value and residual variability of the two compartment model were lower than the one compartment model. The model was parameterized in terms of central volume of distribution (V₁), peripheral volume of distribution (V₂), inter-compartment clearance (Q) and clearance (CL) of teicoplanin. Inter-individual variability was best described by an exponential model and was then estimated for V₁, Q and CL. Residual variability was best described by a combined proportional and additive model.

The allometric size approach was used by incorporating a priori the weight into the basic model (allometric coefficients of 0.75 for CL and Q, 1 for V₁ and V₂), which caused a significant drop in the OFV of 26 points. A further decrease in the OFV of 35.5 units was achieved by implementing creatinine clearance on CL. After implication of the two covariates, the inter-individual variability of CL decreased from 52.2 to 31.8%. Size explained 31.8% and renal function 7.3% of teicoplanin CL variability. The η-shrinkage was 17% for CL. Therefore, the influence of covariates on CL was retained in the model as follows:

5

where CL_i is the CL of the ith individual, WT_i and CL_cr,i are the weight and creatinine clearance of the ith individual, and WT_ref and CL_cr,ref are the reference weight and creatinine clearance values, respectively. In our study, the reference weight and creatinine clearance were the median values of our population, 27.1 kg and 179 ml min^–1. The final population pharmacokinetic parameters are given in Table2.

Table 2.

Population pharmacokinetic parameters of teicoplanin and bootstrap results

	PK parameters	RSE (%)	Bootstrap*
			Median	2.5th	97.5th
V1 (l)
V1 = θ1 × (Bodyweight/27.1)
θ1	12.9	24.7	13.4	6.3	26.7
V2 (l)
V2 = θ2 × (Bodyweight/27.1)
θ2	25.2	19.2	26.5	13.7	44.3
Q (l h–1)
Q = θ3 × (Bodyweight/27.1)0.75
θ3	0.341	25.8	0.317	0.165	0.588
CL (l h–1)
CL = θ4 × (Bodyweight/27.1)0.75 × RF
θ4	0.491	10.1	0.487	0.398	0.608
RF = (CLcr/179) θ5
θ5	0.606	25.7	0.629	0.377	1.048
Inter-individual variability (%)
V1	22.2	110.5	25.3	4.0	69.3
Q	131.5	46.8	146.2	78.6	281.8
CL	31.8	31.9	30.0	17.6	39.8
Residual variability
Proportional (%)	14.8	33.8	14.3	3.0	18.0
Additive (mg l–1)	1.1	68.3	1.3	0.1	3.4

(n = 500)

V₁, central volume of distribution; V₂, peripheral volume of distribution; Q, inter-compartment clearance; CL, clearance; RF, renal function; CL_cr, creatinine clearance;

^*70% successful bootstrap cycles.

Model diagnostics showed acceptable goodness-of-fit criteria for the final model. As shown in Figure1A–D, population and individual predictions are acceptable. In addition, the mean parameter estimates resulting from the bootstrap procedure very closely agreed with the respective values from the final population model, indicating that the final model is stable and can re-determine the estimates of population pharmacokinetic parameters. The results of 500 bootstrap replicates are summarized in Table2. The NPDEs are presented in Figure1E–F. NPDE distribution and histogram met well the theoretical N (0, 1) distribution and density, indicating a good fit of the model to the individual data (Figure1E and F). The mean and variance of NPDE were 0.01 (Wilcoxon signed rank test P = 0.85) and 1.13 (Fisher variance test 0.28), respectively. The pcVPC is shown in Figure1G. The prediction-corrected observed concentrations fit well with the simulated concentrations, confirming the predictive performance of the developed model.

Figure 1.

Model evaluation for teicoplanin. (A and B) Routine diagnostic goodness-of-fit plots: population predicted (PRED) vs. observed concentrations (DV) and individual predicted (IPRED) vs. observed concentrations (DV). (C and D) Conditional weighted residuals (CWRES) vs. time and conditional weighted residuals (CWRES) vs. population predicted concentrations (PRED). (E and F) Normalized prediction distribution errors (NPDE): Q-Q plot of the distribution of the NPDE vs. the theoretical N(0, 1) distribution and a histogram of the distribution of the 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 and semi-transparent grey field represents simulation-based 95% confidence intervals for the median. The observed 5% and 95% percentiles are presented with dashed lines and the 95% intervals for the model-predicted percentiles are shown as corresponding semi-transparent grey fields

Figure2 shows the relationship between individual teicoplanin CL and covariates (body weight and creatinine clearance). The medians of CLs were 0.028, 0.019 and 0.015 l h^–1 kg^–1 for infants, children and adolescents, respectively.

Figure 2.

The relationship between teicoplanin clearance and covariates. (A) Teicoplanin clearance vs. weight, (B) teicoplanin clearance (normalized by size) vs. creatinine clearance. Size is defined as (body weight/27.1)^0.75. Dashed lines depict the typical covariate-CL relationship and (C) teicoplanin clearance (normalized by weight) vs. age

The performance of the developed model was further evaluated in an independent group of 15 children with HM with a mean age of 6.7 years (SD 4.2 years, range 1.4 to 16.3 years), a mean weight of 25.5 kg (SD 18.6 kg, range 11.0 to 81.7 kg), and a mean creatinine clearance of 183.1 ml min^–1 (SD 81.1 ml min^–1, range 77.2 to 306.7 ml min^–1; Schwartz formula). Fifteen concentrations were available and ranged from 5.7 to 30.3 mg l^–1. The Bayesian estimated concentrations were highly correlated with measured concentrations (r² = 0.99). The mean PE and APE were 0.7% (5^th to 95^th percentile, −6.1% to 9.8%) and 5.3%, respectively, indicating a good predictive performance of the developed model on new patients.

Dosing optimization based on pharmacokinetic model

1. Traditional paediatric dose (mg kg^–1 basis). The target attainment rates as a function of dose and age groups for an AUC target of 750 mg l^–1h are shown in Figure3. The kg kg^–1 results in only 5% of infants, 27% of children and 31% of adolescents achieving the target AUC. These simulated values were in agreement with the observed values in the present study. To reach the target AUC of 750 mg l^–1 h in about 50% of patients, 18 mg kg^–1 was required for infants, 14 mg kg^–1 for infants and 12 mgkg^–1 for adolescents (Figure3). The overall risks of underdosing (C_ss,min<10 mg l^–1) and overdosing (C_ss,min>30 mg l^–1) were 13% and 29%, respectively.

2. Patient-tailored dose. A patient-tailored dose was calculated for each patient based on equations 4 and 6:

Figure 3.

Target attainment rates. Target attainment rates for the 100 simulated trials are presented as a function of dose and age group. The AUC target is 750 mg l^–1 h. infants , children , adolescents

6

where bodyweight is in kg and creatinine clearance is in ml min^–1 (Schwartz formula).

The expected AUC(0,24 h) in simulated trials for patients receiving the patient-tailored dose showed a similar median value (749 mg l^–1 h vs. 762 mg l^–1 h), but less variability (5^th–95^th: 446–1272 mg l^–1 h vs. 395–1635 mg l^–1 h) in comparison to the mg kg^–1 basis dosing regimen. The proportion of patients achieving the target C_ss,min (10 to 30 mg l^–1) is 68% using the patient-tailored dose, which is higher than the percentage for the traditional mg kg^–1 basis dose (58%). The proportions of patients with risks of underdosing (C_ss,min <10 mg l^–1) or overdosing (C_ss,min >30 mg l^–1) are 9% and 22%, respectively, using the patient-tailored dose, which is lower than values for the traditional mg kg^–1 basis dose (13% for underdosing and 29% for overdosing) (Figure4) (Table3).

Figure 4.

C_ss,min distribution. Simulated teicoplanin C_ss,min distribution in patients receiving mg kg^–1 basis dose and patient-tailored dose. C_ss,min (mg l⁻¹): <10 , 10–30 , >30

Table 3.

Simulated AUC and C_ss,min for proposed dosing regimens

	Median AUC(0,24 h) (mg l–1 h)	AUC(0,24 h) 5th–95th (mg l–1 h)	Css,min <10 mg l–1	Css,min >30 mg l–1
Traditional paediatric dose (mg kg–1 basis)	762	395–1635	13%	29%
Patient-tailored dose	749	446–1272	9%	22%

Relationship between AUC(0,24 h) and C_ss,min

1. Traditional paediatric dose (mg kg^–1 basis). The medians (5^th–95^th) of C_ss,min obtained with the traditional paediatric dose (mg kg^–1 basis)were 18.8 (5.8–44.0), 21.6 (6.9–61.3), 21.5 (7.2–51.4) mg l^–1 for infants, children and adolescents, respectively. The correlation coefficients (r, Pearson correlation) between C_ss,min and AUC(0,24 h) were 0.92 (P < 0.01), 0.95 (P < 0.01), 0.95 (P < 0.01) for infants, children and adolescents, respectively.

2. Patient-tailored dose. The median (5^th–95^th) of C_ss,min obtained with the patient-tailored dose were 20.9 (8.0–43.7) mg l^–1. The correlation coefficient (r, Pearson correlation) between C_ss,min and AUC(0,24 h) was 0.89 (P < 0.01).

Discussion

In the present study, the population pharmacokinetics of teicoplanin were evaluated for the first time in children with HM. We have shown that similarly to vancomycin, teicoplanin is eliminated with higher clearance than that expected from already available data in the general paediatric population. Pharmacokinetic data are indispensable to evaluate a higher dose of teicoplanin in children with HM.

According to regulatory guidelines, antimicrobial agents are good examples of drugs for which modelling and simulation techniques can be used to develop dosage recommendations in children. The pharmacokinetic–pharmacodynamic breakpoint of teicoplanin was defined as a target AUC(0,24 h) of 750 mg l^–1 h in adults with MRSA infections. Kanazawa et al. showed that an increase in microbiological eradication probability was dependent on AUC and a target AUC of 750 mg l^–1 h gave about 0.9 probability of MRSA eradication, in which the MIC for the isolates was <2 mg l^–1 15. This pharmacokinetic–pharmacodynamic breakpoint is postulated to be similar in children and is therefore used for dosing optimization in children. The current recommended paediatric dose of teicoplanin resulted in a high risk of underdosing in our HM population, as 41% of patients had C_ss,min below the recommended value of 10 mg l^–1. It should be noted that although a C_ss,min of 10 mg l^–1 is generally accepted as the target for most bacteraemia, a higher C_ss,min of 20 mg l^–1 is currently recommended for MRSA 16,17. Our simulation results supported this increased target C_ss,min value. To achieve the target AUC(0,24 h) of 750 mg l^–1 h, the corresponding C_ss,min was about 21 mg l^–1. The significant correlation between AUC and C_ss,min will facilitate TDM, as C_ss,min is much easier to be routinely used to monitor teicoplanin than AUC does. The risk of adverse events can be decreased with the use of TDM and model-based dosage adaptation. Based on the results of the present study, the AUC-based dosage adaptation can be routinely applied using the population pharmacokinetic model.

Patients with malignancy represent a critical population in whom inadequate empirical antibacterial therapy may result in limited efficacy either related to underdosing (infection-related morbidity and mortality) or with overdosing (increased toxicity). In addition, underdosing has a major ‘public health impact’ contributing to the increasing rate of acquired bacterial resistance 18. Population pharmacokinetic modeling demonstrated that teicoplanin pharmacokinetics were well described by a two compartment model with first order elimination. Our results are consistent with the findings of previous population pharmacokinetic studies conducted in adults 19–21. All these three studies in adults have shown that the population pharmacokinetics of teicoplanin were well described by a two compartment model and the elimination of teicoplainin was significantly influenced by renal function. Our findings are also in agreement with those of a pharmacokinetic study in critically ill children, showing that 54% of patients had C_ss,min<10 mg l^–1 with the current dose regimen 22. There are very limited data in neonates. Tarral et al. found a mean clearance and a mean volume of distribution of 0.016 l h^–1 kg^–1 and 0.6 l kg^–1, respectively, in four neonates 23. Based on our data and in order to increase the likelihood of treatment ‘success’, teicoplanin dosage needs to be increased in children with HM.

The major barriers to paediatric pharmacokinetic studies are the relatively large volumes of blood loss during the study period, difficulty in timing of pharmacokinetic samples due to the critical clinical condition and a relatively low rate of informed parental consent 24. The early population pharmacokinetic studies based only on TDM samples provided a practical solution, although, due to the missing pharmacokinetic samples in the distribution phase, only a one compartment model could be built and the estimation of volume of distribution was imprecise. We used opportunistic samples, which were collected in the laboratory from the excess blood taken during routine clinical care, and combined them with TDM samples to perform a population pharmacokinetic analysis in children. By its very nature, an opportunistic sample strategy produces an opportunity to optimize the use of virtually every clinically obtained blood sample by keeping the limited volume of blood remaining after the required laboratory tests, in order to determine drug concentration with sensitive analytical methods adapted to very limited volumes, then to perform and/or enrich pharmacokinetic evaluations. The opportunistic sampling strategy has been successfully incorporated in our recent pharmacokinetic study of ciprofloxacin in neonates and showed similar results compared with the traditional pharmacokinetic design 25,26. The clinical utility of the opportunistic sampling design is undoubtedly in paediatric pharmacokinetic studies. In the present study, the median estimated volume of distribution at steady-state was 1.1 l kg^–1, which was consistent with previous pharmacokinetic studies of teicoplanin in children using rich pharmacokinetic sampling design and non-compartment pharmacokinetic analysis 5,6.

As teicoplainin is almost exclusively eliminated by the renal route, both renal function and size should have important influences on the dosing regimen in children. Covariate analysis fitted well the pharmacokinetic characteristics of teicoplanin and identified that size (expressed as allometric scaling of weight) and creatinine clearance significantly influenced the teicoplanin clearance. The difficulty of incorporation of these covariates in paediatric dose individualization is mainly due to the non-linear correlation between bodyweight, developmental changes and biological/clinical condition in paediatric dosing, as demonstrated in the present study. Clearance, expressed per kg of body weight, is commonly used in paediatric dosing calculations. It has a larger value in children than adults and shows highest value in infants 27. As demonstrated in Figure2C, the clearance decreased with increasing age and, thereby, the uniform weight-based dosing (mg kg^–1) is not adapted to the whole paediatric range. Modelling and simulation approaches quantified the effects of size and renal function on teicoplanin dosing and provided ‘evidence-based’ personalized therapy. We compared the patient-tailored dose with the traditional mg kg^–1 basis dose strategies. The simulation clearly supports the use of a patient-tailored dose, which showed a narrow spread between AUC and C_ss,min values compared with the traditional mg kg^–1 basis dose. The patient-tailored dose resulted in a higher proportion of patients within the target concentrations, associated with lower risk of underdosing or overdosing. This result was in consistent with our previous findings of vancomycin in both neonates 28 and children with malignant haematological disease 8.

The major physiologic and developmental processes during childhood produce well-known dynamic changes in drug pharmacokinetics and disposition. The current policy for drug evaluation in children has emphasized the importance of age on the dosing regimen but the impact of disease, sometimes specific to paediatric patients, has drawn less attention. The dose evaluation studies of antimicrobials are usually based on a ‘non-selected’ paediatric population and do not take into account the potential impact of the disease and/or disease state, which are the main factors that ultimately influence drug pharmacokinetics 29,30. The altered pharmacokinetic parameters have been widely demonstrated in adult critically ill, burn and cancer patients but limited data are available in children. In our previous study of vancomycin, we have shown that an increased vancomycin clearance, thereby an increased dose, was required to reach a similar exposure in children with HM in comparison with the general population 9. Similar results were obtained in the present study with teicoplanin, providing additional data to prove that the disease, disease state and/or factors related to the disease (therapeutic procedures, associated therapies etc.) have a significant impact on pharmacokinetics, thereby requiring dosage adaptation to avoid the risk of underdosing in children with HM. In the present study, the higher teicoplanin clearance was associated with a higher creatinine clearance than that expected in paediatric patients of similar ages. This is most probably related to high glomerular filtration secondary to hyperhydration, which is part of the HM protocol. Similarly to teicoplanin and vancomycin, such a mechanism will impact on the elimination of all renal eliminated drugs.

A limitation of our study is the absence of clinical data to determine response to treatment. Indeed, evaluation of efficacy and safety of antibiotics is required in the target paediatric population. Obviously, the optimal dosing regimen obtained from a pharmacokinetic study might serve as a stepping stone to evaluate clinical response and validate the optimal dosage regimen of teicoplanin proposed here.

In conclusion, in the present study, we developed a population pharmacokinetic model of teicoplanin in children with malignant haematological disease, with weight and creatinine clearance being significant covariates of teicoplanin clearance. The current dosing regimen (10 mg kg^–1) is associated with a high risk of underdosing in this particular group of patients. In order to reach the target AUC of 750 mg l^–1 h, 18 mg kg^–1 was required for infants, 14 mg kg^–1 for children and 12 mg kg^–1 for adolescents. We further developed a patient-tailored dose, which reduced variability in teicoplanin AUC and C_ss,min values compared with the mg kg^–1 basis dose, making the modelling approach an important tool for dosing individualization. A prospective study is warranted to evaluate the potential clinical benefits and safety of this optimized dosing regimen.

Competing Interests

All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare this work was supported by the GRiP (Global Research in Paediatrics, European Commission FP7 project, grant agreement number 261060) and ‘The Fundamental Research Funds of Shandong University’. There are no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years and no other relationships or activities that could appear to have influenced the submitted work.

We acknowledge the technicians (Michel Popon, Jeremie Touati, Samira Benakouche and Yves Médard) for technical support.