Dose evaluation of lamivudine in human immunodeficiency virus‐infected children aged 5 months to 18 years based on a population pharmacokinetic analysis

Abstract

Aim

The objectives of this study were to characterize age‐related changes in lamivudine pharmacokinetics in children and evaluate lamivudine exposure, followed by dose recommendations for subgroups in which target steady state area under the daily plasma concentration–time curve (AUC_0–24h) is not reached.

Methods

Population pharmacokinetic modelling was performed in NONMEM using data from two model‐building datasets and two external datasets [n = 180 (age 0.4–18 years, body weight 3.4–60.5 kg); 2061 samples (median 12 per child); daily oral dose 60–300 mg (3.9–17.6 mg kg^–1)]. Steady state AUC_0–24h was calculated per individual (adult target 8.9 mg·h l^–1).

Results

A two‐compartment model with sequential zero order and first order absorption best described the data. Apparent clearance and central volume of distribution (% RSE) were 13.2 l h^–1 (4.2%) and 38.9 l (7.0%) for a median individual of 16.6 kg, respectively. Bodyweight was identified as covariate on apparent clearance and volume of distribution using power functions (exponents 0.506 (20.2%) and 0.489 (32.3%), respectively). The external evaluation supported the predictive ability of the final model. In 94.5% and 35.8% of the children with a body weight >14 kg and <14 kg, respectively, the target AUC_0–24h was reached.

Conclusion

Bodyweight best predicted the developmental changes in apparent lamivudine clearance and volume of distribution. For children aged 5 months–18 years with a body weight <14 kg, the dose should be increased from 8 to 10 mg kg^–1 day^–1 if the adult target for AUC_0–24h is aimed for. In order to identify whether bodyweight influences bioavailability, clearance and/or volume of distribution, future analysis including data on intravenously administered lamivudine is needed.

What is Already Known about this Subject

• Lamivudine is widely used in human immunodeficiency virus‐infected children.

• Small populations or populations with narrow age ranges have been used before to investigate the pharmacokinetics of lamivudine.

• Dosing in children should be guided on the basis of understanding the developmental changes in pharmacokinetics of drugs.

What this Study Adds

• Bodyweight best explained age‐related pharmacokinetic variability.

• In 94.5% and 35.8% of the children weighing > and <14 kg adult target steady state AUC_0–24h of 8.9 mg·h l^–1 was reached.

• This study indicates that daily dose should be increased to 10 mg kg^–1 for children weighing <14 kg to reach the target area under the daily plasma concentration–time curve.

Table of Links

LIGANDS
Zidovudine

This Table lists key ligands in this article that are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 1.

Introduction

Lamivudine is a nucleoside reverse transcriptase inhibitor (NRTI) that is widely used as part of the treatment of human immunodeficiency virus (HIV)‐infected children. It is currently recommended as part of first‐line NRTI backbone together with either a protease inhibitor or non‐nucleoside reverse transcriptase inhibitor 2. Lamivudine is available in solid and liquid dosage forms, as well as in single entity and fixed dose combination products. Currently, a daily dose of 8 mg/kg is recommended for children, independent of their age and bodyweight 2, 3. Several issues have been raised concerning the treatment with lamivudine, such as bioavailability issues 3, 4, 5, 6, 7 and possible underdosing in the youngest age group 8, 9, 10.

There have been several studies on the pharmacokinetics of lamivudine in children. Most studies applied noncompartmental analysis 5, 6, 9, 10, 11, 12, 13, 14, 15, 16 and some developed a population pharmacokinetic model 8, 17, 18, 19, 20, 21. Most of these studies were based on a small number of children, narrow age ranges or the relationship between parameters and covariate was fixed a priori. None of the models has been validated externally, in other words the models have not been evaluated in how well they generalize to new data that have not been included in the model‐building dataset.

As highlighted before, dosing in children should be based on the understanding of the developmental changes in the pharmacokinetic and/or pharmacodynamic relation of drugs instead of applying the adult mg/kg dose to children 22, 23, 24. For lamivudine, the area under the daily plasma concentration–time curve (AUC_0–24h) is mainly used as a surrogate for the intracellular exposure to lamivudine triphosphate. In adults, an average steady state AUC_0–24h of 8.9 mg·h l^–1 is reached after administration of a daily dose of 300 mg which is the licensed dose with proven efficacy 3. Also, in children, this value is used as a target for lamivudine exposure 8, 16, 17, 25.

The objectives of this study were to characterize age‐related changes in lamivudine pharmacokinetics in infants, children and adolescents, and to test how well this model can be generalized to new patients not included in the model‐building dataset. Based on the developed population pharmacokinetic model, lamivudine exposure upon currently used dosing recommendations was evaluated and, when necessary, a new dose will be calculated for subgroups in which target AUC_0–24h was not reached.

Methods

Patients and treatment

Model building was based on data from two datasets of in total 85 children using lamivudine twice daily 26. These datasets were chosen for model‐building, because almost the entire paediatric age range was represented and body weight of all of the children was known. The first dataset consisted of 64 children, aged 0.5–14.9 years, who participated in the CHAPAS1 trial 26. CHAPAS1 was an open, randomized controlled phase I/II trial designed to assess the appropriate dosing of and adherence to Triomune. Lamivudine was administered in a fixed‐dose combination tablet of lamivudine, nevirapine and stavudine. Daily dosing was based on acquiring an appropriate nevirapine dose. Daily lamivudine dose varied between 60 and 240 mg (6.3–17.6 mg kg^–1). In the second dataset, 21 children, aged 1.7–18.0 years, were included in whom therapeutic drug monitoring on lopinavir was performed as part of routine clinical care. In the available samples, lamivudine concentrations were also determined. Lamivudine was dosed orally according to the PENTA guideline valid at that time 27. Daily lamivudine dose varied between 80 and 300 mg (5.1–10.5 mg kg^–1) for the children included in this cohort. An overview of the patient characteristics is given in Table 1.

Table 1.

Patient characteristics of the children in the two model‐building datasets and the two datasets used for external validation. Values are expressed as median [range]

	Model building			External validation			Total
	Dataset 1 26	Dataset 2	Total model‐building	Dataset 1a	Dataset 2 9, 15, 16	Total external validation
Number of children	64	21	85	24	77	101	180
Age (years)	7.0 [0.5–14.9]	8.0 [1.7–18.0]	7.2 [0.5–18.0]	8.4 [1.6–17.3]	5.8 [0.4–12.8]	6.0 [0.4–17.3]	6.6 [0.4–18.0]
Bodyweight (kg)	16.3 [3.4–29.0]	27.8 [10.5–59.2]	16.6 [3.4–59.2]	30.2 [11.0–54.1]	17.4 [7.4–60.5]	18.2 [7.4–60.5]	17.0 [3.4–60.5]
Number of doses (mean per child)	64 (1)	21 (1)	85 (1)	88 (3.7)	153 (2.0)	241 (2.4)	326 (1.8)
Number of observations (mean per child)	433 (6.8)	205 (9.8)	638 (7.5)	232 (9.7)	1191 (15.5)	1423 (14.1)	2061 (11.5)
Daily dose (mg)	120 [60–240]	200 [80–300]	120 [60–300]	200 [90–300]	150 [60–300]	160 [60–300]	150 [60–300]
Daily dose (mg kg–1)	8.9 [6.3–17.6]	7.8 [5.1–10.5]	8.5 [5.1–17.6]	7.3 [3.9–10.2]	8.6 [4.9–15.0]	8.2 [3.9–15.0]	8.2 [3.9–17.6]
Formulation	FDC: 64	NA: 21	FDC: 64 NA: 21	Solution: 17 Tablet: 9 NA: 62	Solution: 63 Tablet: 90	Solution: 80 Tablet: 99 NA: 62	FDC: 64 Solution: 80 Tablet: 99 NA: 83

^a 16 of the 24 children in this dataset were part of the RONDO trial 28; six of the 24 children were also included in model‐building dataset 2 on a different occasion; FDC: fixed dose combination tablet; NA: not available

External evaluation of the model was performed with two external datasets 9, 15, 16, 28. The first external dataset consisted of 24 children, aged 1.6–17.3 years, in whom therapeutic drug monitoring on lopinavir was performed. Lamivudine concentrations were determined on samples from different occasions (range: 1–10 occasions). Sixteen of these 24 children participated in the RONDO trial 28. Lamivudine was dosed both once and twice daily according to the PENTA guideline valid at that time and daily dose varied between 90 and 300 mg (3.9–10.2 mg kg^–1) 27. The second external dataset consisted of 77 children, aged 0.4–12.8 years, who were included in three studies: PENTA13 9, PENTA15 16 and ARROW 15. All three studies were cross‐over studies to compare the pharmacokinetics of once daily lamivudine dosing vs. twice daily dosing 9, 15, 16. Daily dose varied between 60 and 300 mg (4.9–15.0 mg kg^–1). An overview of the patient characteristics is given in Table 1. Data on different dosing occasions of six children were included in both the model‐building dataset as well as in the dataset for external validation.

The study protocols of the included studies were approved by medical ethical committees and regulatory bodies in each participating country and clinical site. All studies were performed in full conformance with the principles of the Declaration of Helsinki. All the patients were only included if written informed consent was given by the parents.

Blood sampling and assay

For all children included in the model‐building datasets and the second external dataset, at least one complete concentration–time profile after dosing was available (≥6 samples). This also applied for 15 children (63%) included in the first external dataset. Sampling was performed until the end of the dosing interval.

For the data from the CHAPAS1 trial, PENTA trials and therapeutic drug monitoring study, lamivudine concentrations were measured using a high‐performance liquid chromatography assay with ultraviolet detection 29. The lower limit of quantification was 0.05 mg l^–1. As only one value from the model‐building datasets and 12 values from external validation datasets were below the limit of quantification, these values were excluded from the analysis (M1 method 30). For the data from the ARROW trial, lamivudine was measured using a high‐performance liquid chromatography assay with tandem mass spectrometry detection. The lower limit of quantification was 0.0025 mg l^–1.

Pharmacokinetic analysis and model evaluation

Model building was performed in four different steps: 1) testing of both a one‐ and two‐compartment model and different absorption models in order to select a structural model; 2) selection of a statistical model; 3) covariate analysis; and 4) model evaluation. For oral absorption, both a zero order and first order model, a lag time model 31, transit compartment model 31 and combined absorption models were evaluated.

Discrimination between structural models was achieved by comparison of the objective function value (OFV) and the degrees of freedom. A decrease in the OFV of more than 3.8 points was considered statistically significant for the structural model (P < 0.05 based on χ² distribution). The goodness‐of‐fit plots (observed vs. both individual‐ and population‐predicted concentrations and conditional weighted residuals vs. both time and population predictions) were evaluated. Improvement of individual plots, confidence intervals of the parameter estimates and the correlation matrix were also assessed.

Covariate analysis

Covariates were plotted against individual posthoc parameter estimates and the weighted residuals to visualize potential relationships. The covariates bodyweight, age, height and formulation were evaluated. Potential covariates were separately implemented in the model, using a linear or power equation (1).

$$P_i=P_p\times {\left(\frac{{\mathit{cov}}_i}{{\mathit{cov}}_{\mathit{\text{median}}}}\right)}^k$$ (1)

where P_i represents the individual parameter estimate of the i^th subject, P_p is the population parameter estimate, cov is the covariate and k is the exponent. K was fixed at 1 for a linear function or estimated for a power function. The framework proposed by Krekels et al. 32 to systematically evaluate the descriptive and predictive performance of a paediatric model was used as a guide to discriminate between different covariate models. A decrease in OFV of at least 7.8 points was applied to evaluate covariates in forward inclusion. In backward deletion, a more stringent P‐value of <0.001 was used (a decrease in OFV of at least 10.83 points). When two or more covariates were found to significantly improve the model, the covariate causing the largest decrease in OFV was kept in the model. In order to be retained in the model, additional covariates had to reduce this OFV further. The clinical relevance of a covariate relationship was also considered 32. In order to confirm the final covariate model, individual and population parameter estimates were plotted against the most predictive covariate to evaluate whether the individual predicted parameters were equally distributed around the population predicted parameters.

Internal evaluation procedure

The final model was evaluated using three methods 32: the bootstrap method for model stability and precision 33, and both the normalized prediction distribution error (NPDE) method 34, 35 of the model parameter estimates and a visual predictive check 36 for model accuracy. The model‐building dataset was resampled to produce 2000 new datasets of the same size, containing a different combination of individuals. The final model was sequentially fitted to all of these newly generated datasets. The parameter estimates were summarized in terms of median values and 95% parametric confidence intervals, and were compared with the estimates obtained from the model‐building datasets.

External evaluation procedure

External evaluation of the model was performed with two external datasets 9, 15, 16, 28, as described above. These datasets were not included when the model was fitted to the data.

The final pharmacokinetic model was used to simulate concentrations for each data point in the two external datasets. Both interindividual and residual variability were included in the simulations. Additionally, the final pharmacokinetic model was used to compute the NPDE 34, 35 for each of the external datasets. Finally, the parameters of the final model were re‐estimated on the basis of the two model‐building datasets combined with both external datasets.

Evaluation of currently used dosing guidelines

Intracellular triphosphate levels of lamivudine have been found to be predictive to anti‐HIV response with the intracellular triphosphate levels being directly related to the lamivudine plasma concentration 37. Target AUC_0–24h was 8.9 mg·h l^–1, which is the AUC_0–24h obtained in adults after once daily administration of 300 mg being the licensed dose for which efficacy has been shown 3. For the first available dosing occasion per individual, AUC_0–24h was noncompartmentally derived from the estimated individual parameter estimates. Based on the results, a dose adaptation was proposed for subgroups of children not reaching the target AUC_0–24h.

Software

The pharmacokinetic analysis and evaluation procedures were performed using the nonlinear mixed‐effects modelling software NONMEM version 7.3 (Icon Development Solutions, Hanover, MD). Tools like PsN version 4.2.0 38 (University of Uppsala, Sweden), Pirana version 2.9.0 (Pirana Software & Consulting BV, Amsterdam, the Netherlands) and R version 3.1.1 (R Foundation for Statistical Computing, Vienna, Austria) were used to visualize and evaluate the models. For the NPDE analysis, the NPDE software package in R 35 was used.

Results

Population pharmacokinetic model‐building

Model building was based on 638 observations from 85 children while external evaluation was based on 1423 observations from 101 children (Table 1). A two‐compartment model with sequential zero order and first order absorption best described the data. A two‐compartment model was preferred over a one‐compartment model, since the model‐building data, and especially the highest concentrations, were more accurately described with a two‐compartment model (dOFV: 140.701 points). The final model was parameterized in terms of a zero‐order absorption phase (D1), which was followed by a first order absorption process (K_a) (i.e. sequential zero order and first order absorption model), apparent clearance (Cl/F), intercompartmental clearance (Q/F), and apparent volumes of distribution of the central compartment (V₂/F) and peripheral compartment (V₃/F). Because there were difficulties estimating reliable values for V₃/F, the model was simplified by equalising V₃/F to V₂/F. The residual variability was best described using a combined additive and proportional error model.

Systematic covariate analysis

The covariate analysis identified bodyweight as the most important covariate for both Cl/F and V₂/F. After implementation of this covariate, OFV decreased with 47.573 points. The exponent for the effect of bodyweight on Cl/F was 0.506 (20.2%) and for the effect on V₂/F 0.489 (32.3%). The parameter estimates for the simple and final model are shown in Table 2. In Figure 1, the individual estimates of variability of Cl/F and V₂/F are plotted against bodyweight for the simple and final models. A significant part of the interindividual variability is explained after inclusion of bodyweight as a covariate, with a decrease of 8.9% in the interindividual variability of Cl/F and 9.5% in the interindividual variability of V₂/F (Table 2). After inclusion of bodyweight, no other covariates (i.e. age, height or formulation) could be identified (P > 0.05). Estimate of variability of shrinkage was 4.13%, 6.25% and 23.1% for respectively Cl/F, V₂/F and D1.

Table 2.

Population parameter estimates of the i) simple and ii) final pharmacokinetic model based on two model‐building datasets, iii) the values obtained after bootstrap of the final pharmacokinetic model, and iv) the parameter estimates after combining the model‐building data with the two external datasets

Parameter	Simple pharmacokinetic model	Final pharmacokinetic model	Bootstrap final pharmacokinetic model	Model building data and external data
Fixed effects
Cl/F16.6 kg (l h–1)	13.2 [4.8]	13.2 [4.2]	13.2 [12.1–14.3]	13.1 [2.8]
θ in Cl/F16.6 kg × (BW/16.6)θ		0.506 [20.2]	0.524 [0.298–0.716]	0.372 [18.6]
V2/F16.6 kg = V3/F16.6 kg (l)	37.0 [7.6]	38.9 [7.0]	37.8 [30.3–43.9]	36.6 [6.8]
θ in V2/F16.6 kg × (BW/16.6)θ		0.489 [32.3]	0.531 [0.127–0.836]	0.581 [19.8]
Q/F (l h–1)	2.09 [17.7]	2.02 [13.4]	2.14 [1.53–3.76]	2.65 [11.3]
D1 (h)	0.697 [15.1]	0.847 [10.3]	0.823 [0.493–1.05]	0.655 [9.7]
Ka (h−1)	2.47 [11.5]	3.41 [17.3]	3.16 [1.56–6.78]	1.73 [12.1]
Interindividual variability
ω (Cl/F)	46.8 [7.5]	37.9 [7.9]	37.4 [0.319–0.436]	33.6 [6.4]
ω (V2/F)	64.5 [6.9]	55.0 [9.0]	55.9 [0.467–0.686]	43.5 [7.2]
Omega block (Cl/F – V2/F)	0.848	0.820	0.812 [0.797–0.813]	0.718
ω (D1)	81.5 [10.7]	70.4 [10.5]	72.1 [0.560–0.986]	79.6 [9.0]
Residual variability
σ2 (proportional)	13.3 [9.2]	14.7 [9.3]	14.4 [0.118–0.174]	30.1 [5.2]
σ2 (additive)	0.0477 [10.9]	0.0450 [11.4]	0.0443 [0.0340–0.0542]	0.088 [27.9]

Data presented as value [%RSE]; bootstrap results presented as median [95% CI]

θ: parameter of interest; ω, σ²: variance, expressed as %CV, BW: bodyweight; CI: confidence interval; Cl/F_{16.6 kg}: apparent clearance for a typical individual with BW of 16.6 kg; D1: duration of zero order absorption; K_a: rate constant of first order absorption; Q/F: intercompartmental clearance; V₂/F_{16.6 kg}: volume of distribution of the central compartment for a typical individual with BW of 16.6 kg; V₃/F_{16.6 kg}: volume of distribution of the peripheral compartment for a typical individual with BW of 16.6 kg

Figure 1.

Interindividual variability (dots) for the simple (left) and final model (right) for apparent clearance [estimate of variability (ETA) on apparent clearance (Cl/F)] and apparent volume of distribution [ETA on volume of distribution of the central compartment (V₂/F)] vs. weight (two model‐building datasets) with trendline (blue line) and 95% confidence interval (grey area)

Internal evaluation of the final pharmacokinetic model

Table 2 gives an overview of the parameter estimates of the simple and final model, together with the values obtained from the bootstrap analysis. The median estimated values based on the bootstrap were within 10% of the values obtained in the final model. In Figure 2A,B,C,D, the goodness‐of‐fit plots, observed vs. individual‐ and population predicted concentrations and conditional weighted residuals vs. population predicted concentrations and time, are given for the final model, while in Figure 2E a histogram of the NPDE is shown. No trend was seen in the NPDE vs. time or vs. predicted concentrations (results not shown). A visual predictive check shows the adequacy of the model (Figure 3). Figure S1 shows that the data in different weight groups are well described.

Figure 2.

Observed vs. individual predicted concentrations and observed vs. population predicted concentrations of (A,B) the two model‐building datasets, (F,G) external dataset 1 and (I,J) external dataset 2. Conditional weighted residuals (CWRES) vs. population predicted concentrations (C) and time (D) for the model‐building datasets are shown. The histograms show the distribution of the normalized prediction distribution error (NPDE) of the (E) model‐building datasets, (H) external dataset 1 and (K) external dataset 2

Figure 3.

Visual predictive check of the final pharmacokinetic model. The dots represent observed concentrations, the solid line represents the median of the simulated profile and the dotted lines represent the 95% confidence interval. The grey shaded areas are the 95% confidence intervals for the same percentiles of the simulations

External evaluation of the final pharmacokinetic model

The predictive performance of the final model was evaluated using two external datasets 9, 15, 16, 28 (Table 1). In Figure 2, observed vs. individual predicted concentrations (Figure 2F,I) and observed vs. population‐predicted concentrations (Figure 2G,J) are given for both external datasets. Additionally, the histograms of the NPDE are shown in Figure 2H,K. While the final model is able to predict the data in external dataset 2 without bias, a slight bias is seen for external dataset 1, in which the sampling was more sparse, compared to the model‐building datasets. This bias is observed in Figure 2G, which shows observed vs. population predicted concentrations, as well as in Figure 2H.

When the parameters of the final model were re‐estimated on the basis of the model‐building datasets and external datasets, fairly similar parameter values were obtained (Table 2). The concentrations in all four datasets could be well described by this model without bias and with adequate precision (Figure S2).

Evaluation of currently used dosing guidelines

Of the children with a bodyweight >14 kg, 94.5% reached the adult target AUC_0–24h of 8.9 mg·h l^–1 with the currently administered daily dose. However, this did not hold for all children with a bodyweight <14 kg (Figure 4). If the daily dosage for these children is increased to at least 10 mg kg^–1 day^–1, it is expected that most children will have adequate exposure to lamivudine (64.2% before dose adaptation, 92.5% thereafter; Figure 5).

Figure 4.

Simulated area under the daily plasma concentration–time curve (AUC_0–24h) vs. daily dose administered (mg) split by bodyweight: ≤14 kg, 14–21 kg, 21–30 kg and >30 kg. The dotted line indicates an AUC_0–24h of 8.9 mg·h l^–1 (adult target for once daily dosing 3). Vertically occurring sequences of dots occur because of fixed dose tablets

Figure 5.

Simulated area under the daily plasma concentration–time curve (AUC_0–24h) vs. daily dose administered (mg/kg bodyweight) for children with a bodyweight ≤14 kg after administered dose (A) and adapted dose where a minimum of 10 mg/kg is administered (B). The dotted line indicates an AUC_0–24h of 8.9 mg·h l^–1 (adult target for once daily dosing 3)

Discussion

A model‐based approach has been applied in order to describe the pharmacokinetics of lamivudine in children. The model was based on a large and relatively rich dataset, since for most of the children at least six samples within one dosing interval were available. Also, the full paediatric age range is covered, with a large proportion of children below the age of 3 years [n = 16 (18.8%) in the dataset used for model‐building; n = 41 (22.8%) in total]. The two‐compartment model used to describe the data is in agreement with previously developed models 8, 17, 21, 39. In order to obtain reliable estimates of the peripheral volume, the model had to be simplified by stating that the central and peripheral volume of distribution were equal to each other (Table 2); however, this did not lead to reduced descriptive or predictive values (Figures 2 and S1).

Bodyweight best predicted the developmental changes in both apparent lamivudine clearance and apparent central volume of distribution. This is consistent with previous studies 8, 17, 18, 19, 20, 21. Although the typical parameter estimates for an individual of 16.6 kg are comparable, the estimations of both scaling exponents were lower in our analysis 8, 17, 18, 19, 20, 21. Remarkable is the difference in the relationship between apparent volume of distribution and bodyweight. The majority of the performed studies fixed this relationship a priori to 1 8, 17, 19, 20, 21. Piana et al. found an exponent of 0.635 18, which seems very close to the exponent of 0.489 we identified (Table 2). In this respect, we emphasize that the exponent we identified is the exponent for Cl/F or for V₂/F and since no data after intravenous administration of lamivudine were available, we cannot distinguish between the influence of bodyweight on bioavailability, clearance and/or volume of distribution.

The stability of the final model was indicated by the NPDE and the bootstrap as well as the ability to predict external dataset 2 accurately. For external dataset 1, the predictive performance was somewhat biased. This may be explained by the (sparse) nature of the data available in that dataset. The data were derived from routine clinical care, where data collection is probably less accurate. When the data from all datasets were combined and analysed together, the data in external dataset 1 were described without any bias (Figure S2).

The target AUC_0–24h of 8.9 mg·h l^–1 that has been identified in adults was reached in 85.6% of the children. However, 35.8% of the children with a bodyweight below 14 kg did not reach this target. It was shown previously that lamivudine exposure was lower in the youngest group of children compared to older and heavier children 8, 9, 10, 40. For children older than 5 months and with a bodyweight below 14 kg we calculated that the target AUC_0–24h can be reached with a dose of at least 10 mg kg^–1 day^–1, based on the expected apparent clearance. The same dose was also proposed by Bouazza et al. for children with a body weight <17 kg 8. We chose our cut‐off body weight in line with the approved dosing regimens of the antiretroviral drugs frequently used in children 2, 41 and with the data found in our model. The relativeness of this cut‐off point and potential considerations for paediatric fixed dose combinations has already been described before 42.

In the model, the absorption phase was relatively difficult to describe, which can partly be explained by the limited data available for that part of the concentration–time profile. In most paediatric population pharmacokinetic models for lamivudine, a first‐order absorption model is used 8, 17, 19, 20, 39. However, also a delay in absorption with either a lag‐time 18 or transit compartments 21 has been described. During model‐building, all these absorption models, as well as a zero‐order absorption, were tested. A sequential zero‐ and first‐order absorption model was finally found to best describe the data (Table 2).

A limitation of the current study is that we could not fully study the influence of the formulation on the pharmacokinetics. As shown before, the type of drug formulation can affect the lamivudine exposure significantly in children 5, 6, 20, 25. However, similar to previous population pharmacokinetic studies 17, 18, 43, the formulation used by the children could not be identified as a possible covariate in our study as information on the formulation used was not complete for all of the children (Table 1). Next to this, we could not study the influence of renal function in this analysis. Lamivudine is a renally excreted drug 3, 4 and it has been shown in adults that renal function can affect the pharmacokinetics of lamivudine 4, 44, 45, 46. Even though in several paediatric studies, serum creatinine could not be identified as possible covariate for clearance 17, 19, 20, we could not study this covariate as information on renal function was incomplete.

In conclusion, lamivudine pharmacokinetics were best described by a two‐compartment model with sequential zero‐order and first‐order absorption. Bodyweight was found as covariate on apparent clearance and apparent central volume of distribution, both in a power function (exponent of respectively 0.506 and 0.489). The model can be generalized to patients not included in the model‐building dataset. In order to identify whether these (nonlinear) changes result from changes in bioavailability, clearance and/or volume of distribution, future analysis, which includes intravenously administered lamivudine, is warranted. The results of this study suggest that the currently recommended dose for children aged 5 months to 18 years and with a body weight below 14 kg should be increased to at least 10 mg kg⁻¹ day⁻¹ in order to reach an AUC_0–24h of 8.9 mg·h l^–1.

Competing Interests

All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: CK had support from NWO and TI Pharma for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous 2 years; no other relationships or activities that could appear to have influenced the submitted work.

This study was performed within the framework of Top Institute Pharma project number D2‐501. Catherijne Knibbe is supported by the Innovational Research Incentives Scheme (Vidi grant, June 2013) of the Dutch Organization for Scientific Research (NWO). We thank all children, families, and staff from the centres participating in the CHAPAS1, PENTA13, PENTA15, ARROW and RONDO studies.

Supporting information

Figure S1 Diagnostic plots (observed vs. predicted concentrations in the model‐building data) of the final model split by bodyweight: a: <10 kg, b: 10–15 kg, c: 15–20 kg, d: 20–25 kg, e: 25–30 kg, f: >30 kg. Dots indicate the observed concentration vs. population predicted concentration, grey dotted lines show x = y, red striped lines show trend line

Figure S2 Diagnostic plots of the model based on both the two model‐building and the two external datasets: observed vs. population predicted concentrations split by dataset: a: dataset 1, b: dataset 2, c: dataset 3, d: datset 4. Datasets 1 and 2 have been used for model‐building, datset 3 and 4 for external validation. Dots indicate the observed vs. population predicted concentration, grey lines show x = y, red lines show trend line

Janssen, E. J. H. , Bastiaans, D. E. T. , Välitalo, P. A. J. , van Rossum, A. M. C. , Jacqz‐Aigrain, E. , Lyall, H. , Knibbe, C. A. J. , and Burger, D. M. (2017) Dose evaluation of lamivudine in human immunodeficiency virus‐infected children aged 5 months to 18 years based on a population pharmacokinetic analysis. Br J Clin Pharmacol, 83: 1287–1297. doi: 10.1111/bcp.13227.