Population pharmacokinetics of azithromycin and chloroquine in healthy adults and paediatric malaria subjects following oral administration of fixed-dose azithromycin and chloroquine combination tablets

The second study was an open-label, randomized, single dose study to estimate the bioavailability of AZCQ tablets in healthy adult volunteers in the US. Subjects in this study were randomized to receive either two AZCQ tablets (AZ/CQ 250/155 mg strength tablet; test treatment) or co-administration of individual commercial tablets of AZ 500 mg and CQ 300 mg (reference treatment).

In the paediatric study, sparse blood samples were collected from both cohorts to determine AZ concentration in serum and CQ/desethylchloroquine (DECQ) concentration in plasma prior to AZCQ dosing (window: –1 to 0 h) on day 0 and day 2. On day 2, blood samples were also collected 3 h (window: 2–4 h) and 8 h (window: 6–10 h) after AZCQ dosing. In addition, blood samples were collected at a random time point on day 7. Blood samples in the bioavailability study were collected to assess serum AZ and plasma CQ concentrations at 0 (pre dose), 0.5, 1, 2, 3, 4, 6, 8, 12, 16, 24, 36, 48, 72 and 96 h post dose.

PK samples from both studies were analysed in the same bioanalytical laboratories (Centero Research in Houston, TX, USA for CQ and DECQ; Bioanalytical System Ltd in Warwickshire, UK for AZ). Plasma or serum samples were analysed for serum AZ, plasma CQ and plasma DECQ concentrations using validated, sensitive, specific high-performance liquid chromatography with tandem mass spectrometry assays[8]. The assays had a lower limit of quantification of 10.0, 1.0 and 0.50 ng/mL for AZ, CQ and DECQ, respectively. In the data analysis, observations that were below the limit of quantification were excluded from the database. The assay accuracy (expressed as % relative error: %RE) of the quality control samples used during the sample analysis ranged from 0% to 6.1% in the paediatric study and from -2.1% to 4.7% in the adult study for AZ, from -3.2 to 9.8% in the paediatric study and from -0.4% to 3% in the adult study for CQ, and from 0.0% to 14.5% in the paediatric study for DECQ. The assay precision data were ≤ 4.9%CV in the paediatric study and ≤4.7% in the adult study for AZ, ≤11.2% CV in the paediatric study and ≤8.9% in the adult study for CQ, and ≤10.8% in the paediatric study for DECQ.

Concentration-time data collected in both studies were analysed using the mixed-effects modelling tool, NONMEM® (version 7, level 2; ICON Plc, Dublin, Ireland). All clearance and distribution parameters are reported as the ratio of the parameter and bioavailability since oral dosing was not accompanied by an intravenous dose. All evaluations used a log-transform both sides approach.

Population PK models were constructed to evaluate the concentration-time data for AZ and CQ and to identify any covariates predictive of PK behaviour using standard approaches. The structural model consisted of a compartmental PK model to describe the observed concentration-time profiles as well as between subject variability in the PK behaviour. For both AZ and CQ, two- and three-compartment models were evaluated. In addition, a one-compartment model was also evaluated for completeness, but did not describe the data well and was therefore not used for either drug.

The covariate models in this PK model were defined to represent the shift in value of the parameter of interest to a value for a hypothetical patient. The reference patient for this analysis was 40 years of age, weighed 70 kg and had a body surface area of 1.73 m² and a body mass index of 15 kg/m². Covariates examined as potential predictors of PK activity are listed in Table 1. Only those covariates that individually influenced the structural parameters were added to create the full model. The full model then underwent reduction where covariates were removed one at a time and the impact of removal was assessed (backward deletion method). The covariate was retained in the final model if its removal resulted in an increase in the objective function of ≥10.83 points (p < 0.001) from the full model.

Two different allometric models using weight normalized PK parameters to size were evaluated[9]. The first model used an allometric coefficient of 0.75 for clearances and 1 for volumes. The second model used estimated allometric coefficients for clearances and volumes. Wherever possible, symmetric 95% confidence intervals were computed using the asymptotic standard errors of the parameter estimates.

The final dataset consisted of 1,198 and 1,197 evaluable measurable AZ and CQ concentration observations from 219 subjects including 40 adults, 123 paediatric subjects aged < 5 years and 56 paediatric subjects aged 5 to 12 years.

The final model for AZ in serum following AZCQ administration was a three-compartment model with linear clearance and absorption and no absorption lag time. The final model for CQ in plasma was a two-compartment model with linear clearance and an absorption lag time. Both models included actual body weight with fixed allometric scaling. In the AZ model, shrinkage was acceptable (defined as ≤20%) for clearance (CL/F), but was high (51.8%) for central compartment volume of distribution (V1/F) that is considered likely due to the sparse sampling in the paediatric study. In the CQ model, shrinkage was acceptable for CL/F and V1/F, but was elevated for the absorption rate constant (62%) that could be due to the sparse sampling in the paediatric study. The condition number for the base model was less than 20 (AZ: 13.4; CQ: 5.3) for both AZ and CQ suggesting no notable co-linearity. Thus, covariate evaluations on CL/F were considered appropriate.

DECQ only had sparse concentration data from the paediatric patients and no data from the adult bioavailability study were available; thus, a population PK model for DECQ could not be identified.

The parameters for the final AZ and CQ population PK models are provided in Table 2. Body weight was found to be influential for both AZ and CQ pharmacokinetics.

Good concordance was noted between the values predicted from the PK model versus the observed concentrations of AZ and CQ (Figure 1). Goodness of fit plots also indicated that the PK models adequately described the PK data for both AZ and CQ and did not suggest substantial bias (Figure 2). A visual predicted check from all the data in the final model indicated that, for both AZ and CQ, the majority of observed concentrations fell within the predicted interquartile ranges (Figure 3).

There was no overall bias or evidence of substantial model misspecification for either AZ or CQ. Body weight as an allometric model was the only covariate in the final AZ and CQ PK models.

The dose-normalized AUC values from time 0 to infinity (AUC_inf) of AZ and CQ were estimated from the respective administered doses and the individual subject clearances (CL/F) then normalized by the administered dose divided by body weight (AUC_inf/dose = dose/CL/[dose/weight]). The paediatric population had significantly lower values than that of the adult population in the dose (mg/kg)-normalized AUC_inf for both AZ (0.488 vs 0.745 mg*h/L/(mg/kg), p < 0.00001) and CQ (0.836 vs 1.27 mg*h/L/(mg/kg), p < 0.00001), indicating a lower (L/h/kg) clearance value in adults as compared to paediatric subjects. Comparison of dose-normalized AUC_inf for both AZ and CQ by age group is graphically presented in Figure 4.