The blood schizonticidal activity of tafenoquine makes an essential contribution to its prophylactic efficacy in nonimmune subjects at the intended dose (200 mg)

Tafenoquine (TQ) is an 8-aminoquinoline anti-malarial being developed for malaria prophylaxis. It has been generally assumed that TQ, administered prophylactically, acts primarily on the developing exoerythrocytic stages of malaria parasites (causal prophylaxis), and that polymorphisms in metabolic enzymes thought to impact the activity of other 8-aminoquinolines also inhibit this property of TQ. Furthermore, it has been suggested that a diagnostic test for CYP2D6 metabolizer status might be required. In field studies in which metabolic status was not an exclusion criteria, TQ has been shown to exhibit similar prophylactic efficacy as blood schizonticidal drugs (mefloquine). Also, its blood schizonticidal and anti-relapse efficacy is independent of 2D6 metabolizer status. The most reasonable explanation for the field study results, supported by other clinical and non-clinical data, is that TQ is not completely causal and exhibits substantial blood schizonticidal activity at the intended dose. Pharmacokinetic simulations demonstrate that trough concentrations of TQ exceed the proposed MIC of 80 ng/ml in >95% of individuals. Based on these data a companion diagnostic for CP450 enzyme status is not required. Electronic supplementary material The online version of this article (doi:10.1186/s12936-017-1862-4) contains supplementary material, which is available to authorized users.


Study Numbers: 044 and 033
Objectives: The objectives of this analysis were: • to validate the NONMEM input files used in the prior analyses, • to verify a population PK model of tafenoquine in a target population of soldiers on military deployment and to verify the effect of covariates (e.g. body weight, age, sex, phospholipidosis, creatinine clearance, presence of malaria) on the PK characteristics of tafenoquine

Methods:
Study 044: The study was a prospective study of tafenoquine in Thai soldiers on monthly prophylaxis deployed along the Thailand-Cambodian border.
Approximately 135 male Thai soldiers participated in the study and were treated with artesunate (300 mg on Day 1, 120 mg daily on Days 2 and 3) plus 200 mg daily doxycycline for 7 days to remove any pre-existing malarial infection. After pre-treatment, 104 soldiers received 400 mg tafenoquine (free base) daily for 3 days followed by 400 mg tafenoquine monthly for 5 consecutive months.
Study 033: The study was a prospective, randomized, double-blind comparative study of tafenoquine and mefloquine in Australian soldiers on weekly malaria prophylaxis deployed on peacekeeping duties for 6 months in East Timor.
Approximately 490 subjects were given a loading dose of 200 mg tafenoquine (free base) per day for three consecutive days followed by an oral weekly maintenance dose of 200 mg tafenoquine for 6 months.
Population PK analyses was carried out using NONMEM Version 7. A one-compartment PK model with first order absorption and elimination was used to describe the PK of tafenoquine. For study 044, inter-individual and residual variability terms were included in the PK model. For study 033, inter-individual, residual variability and interoccasion variability on clearance terms were included in the PK model. As outlined in the publications, a covariate analysis was performed to identify potential covariates affecting the PK of tafenoquine and to evaluate the extent to which the covariates accounted for the variability in the overall response. The validity of the final PK model was evaluated for study 033 using bootstrapping and the visual predictive check as outlined in the publication.

Pharmacokinetic
Based upon the modeling results in the previous publications of study 044 1 and 033 2 , a onecompartment pharmacokinetic model with first order absorption and elimination was used to describe the tafenoquine concentration versus time profiles. A schema of the onecompartment PK model is displayed in Figure 1-1.

Conclusions:
• For both studies 044 and 033, Population PK parameters of tafenoquine and associated variability using the final model are in good agreement with the published results.

INTRODUCTION
Tafenoquine is an 8-aminoquinoline antimalarial agent active against all stages of the malarial parasite. Given the broad range of efficacy during the lifecycle of the parasite, tafenoquine has potential therapeutic uses as a malarial prophylactic and in relapse prevention.
This population pharmacokinetic (PK) analysis was performed using data obtained from a Phase-II clinical study of tafenoquine in Thai soldiers deployed on security operations on the Thailand-Cambodian border (study 044) 1 and in a Phase-III clinical study of tafenoquine in Australian soldiers deployed for 6 months in areas endemic with malaria (study 033) 2 .

OBJECTIVES
The objectives of this analysis were: • to validate the NONMEM input files used in the prior analyses, • to verify a population PK model of tafenoquine in a target population of soldiers on military deployment and to verify the effect of covariates (e.g. body weight, age, sex, phospholipidosis, creatinine clearance, presence of malaria) on the PK characteristics of tafenoquine

Primary Endpoints
• Endpoints for population PK were the population estimates of PK parameters (e.g., CL, V, Ka), interoccasion variability (IOV, study 033 only), associated inter-subject variability and residual error.
• Endpoints for the covariate analysis were the confirmation of significant covariates that are retained in the PK model and quantification of their effects.

Study Design
Study 044: The study was a prospective study of tafenoquine in Thai soldiers on monthly prophylaxis deployed along the Thailand-Cambodian border.
Approximately 135 male Thai soldiers participated in the study and were treated with artesunate (300 mg on Day 1, 120 mg daily on Days 2 and 3) plus 200 mg daily doxycycline for 7 days to remove any pre-existing malarial infection. After pre-treatment, 104 soldiers received 400 mg tafenoquine (free base) daily for 3 days followed by 400 mg tafenoquine monthly for 5 consecutive months. Study 033: The study was a prospective, randomized, double-blind comparative study of tafenoquine and mefloquine in Australian soldiers on weekly malaria prophylaxis deployed on peacekeeping duties for 6 months in East Timor.
Approximately 490 subjects were given a loading dose of 200 mg tafenoquine (free base) per day for three consecutive days followed by an oral weekly maintenance dose of 200 mg tafenoquine for 6 months.

Pharmacokinetic Collection
Study 044: Blood samples for the determination of tafenoquine concentration in plasma for PK evaluation were taken randomly in the field after beginning the loading dose at approximately 8, 24, 48, 56 hours and then at every 3 to 4 day intervals up until the first monthly dose. After each monthly dose, samples were obtained at approximately 8 hours post-dose, mid-month, and at the end of the dosing interval (trough concentration). Following the last monthly dose, samples were obtained at 4, 8, 12, and 24 hours post dose and at every 3 to 4 day intervals for 2 months.
Study 033: Blood samples for the determination of tafenoquine concentration in plasma for PK evaluation were taken at prerandomized times after the last loading dose and then at prerandomized times at weeks 4, 8 and 16. Samples were collected on predetermined days after dosing on each of the assessment weeks. The predetermined days included Day 1 (early postdose, absorption phase), Days 3 and 5 (72 hours to 120 hours postdose), and Day 7 (predose, trough phase). The date and exact time of blood sampling were recorded.

Description of Input Data
All data used in the population PK analysis were obtained from data used in previously published tafenoquine modeling efforts 1,2 . Approximately 135 male Thai soldiers participated in study 044 and approximately 490 subjects participated in study 033.

Handling of Input Data
The data were prepared for analysis and validated using SAS Version 9.1.3 (SAS Institute Inc., Cary, NC). Actual dosing and actual sampling times were used for the analysis.

Bioanalytical Methodology
For study 044, plasma samples were analyzed for tafenoquine concentration by HPLC with fluorescence detection. For study 033, plasma samples were analyzed for tafenoquine concentrations using a validated LC/MS/MS method.

Data Analysis
Population PK analysis was carried out using NONMEM Version 7.

Study 044:
The tafenoquine data from a prior Phase II 1 (study 044) was used to establish a structural PK model and explore covariates.  Source: Table 2 1 .
An additional model (Model 12) was explored using model 1 and estimating the off diagonal elements of the Covariance Matrix (Omega matrix) using the $OMEGA BLOCK (2) function as mentioned in the text 1 .

Study 033:
The tafenoquine data from a prior Phase III 2 (study 033) was used to establish a structural PK model and explore covariates. Table 4-2 outlines model development:

Description of PK Model
Based upon the modeling results in the previous publications of study 044 1 and 033 2 , a onecompartment pharmacokinetic model with first order absorption and elimination was used to describe the tafenoquine concentration versus time profiles. A schema of the onecompartment PK model is displayed in Figure 4-1. This one-compartment PK model was specified in the NONMEM control file and was parameterized in terms of CL, V, and Ka, using the PREDPP ADVAN2 with the TRANS2 subroutine. First order estimation was used for study 044 and first order conditional estimation (FOCE) with interaction was used as the estimation method for study 033.

Pharmacokinetic
Between-individual variability (ETA or η), which is the difference between the individual parameter estimate and the population mean estimate of final PK parameters, was described by an exponential error model as described here for both studies 044 and 033:

Pharmacokinetic
Residual variability (EPSILON or ε), which is the difference between the observed and the individual predicted observations of the final PK model, was described as an exponential error model for study 044 as shown below:

Pharmacokinetic
Interoccasion variability (IOV, kappa or κ), is a random variable representing the variability of a PK parameter across different occasions where each occasion is defined as a dose (or several sequential doses) followed by at least one observation. IOV was assumed to be normally distributed having a mean of 0 and a variance of π 2 and that the variance of each parameter was from the same distribution. IOV was described on CL/F for study 033 as shown below: where: ij CL is the estimated parameter for the i th individual on the j th occasion.
CL is the population value for the parameter.

Covariate Analysis
Covariate analysis was performed to the base PK model to identify potential covariates and to evaluate the extent to which the covariates accounted for the variability in the PK parameters as outlined in each publication 1,2 and presented in Table 4-1 and Table 4-2.
The following variables were selected in the publication for evaluation as potential covariates of CL/F and V/F for study 044: weight (WT), age, presence or absence of malaria (MAL) and the following were evaluated for study 033: age, creatinine clearance (CLCR), presence or absence of malaria (HIST), sex, and weight (WT). It must be noted that in the publication for study 033, Phospholipidosis (PHOS) was considered as a covariate, however, a corresponding binary variable was not identified in the clinical data nor in the provided NONMEM dataset. The binary variable HIST, which was supplied in the NONMEM dataset was substituted in its place.

USAMMDA
Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final

Model Evaluation
As reported in the reference paper, the final PK model for study 033 1 was evaluated using bootstrapping and a posterior predictive check. The final PK model for study 044 2 was evaluated by plotting observed tafenoquine concentrations versus model predicted concentrations, weighted residuals versus predicted concentrations, and weighted residuals versus subject ID as reported in the publication.

Bootstrapping
For study 033, bootstrapping was reported in the publication 1 . As a result, bootstrapping was performed to evaluate the final PK model. Bootstrapping is data re-sampling method for estimating sampling variances, confidence intervals, and stability of regression models.
Using the bootstrap approach, the bootstrap parameter values are obtained by repeatedly fitting the final population model to a reasonable number of bootstrap samples. The mean and confidence interval (CI) values of the bootstrap parameters are then compared to the final population model parameter estimates and associated CIs from NONMEM.
Bootstrap analysis procedure consisted of the following steps: 1. The replication of the data file is obtained by random draw (with replacement) from the original data file.
2. The final model is fitted to the resulting data file, and the model parameter estimates are saved for the final analysis.

Posterior predictive check
For study 033, a posterior predictive check was reported in publication 1 . As a result, a posterior predictive check was performed by simulating the tafenoquine concentrations from the original datasets using the parameter estimates obtained in the final PK modeling step. One thousand (1000) predicted profiles were simulated for each original subjects/patients. Random effects were included in the simulation. The median, 5 th percentile, and 95 th percentile PK concentrations versus time profiles from the simulations were compared with observed tafenoquine concentrations. Simulated results (5 th percentile, median, and 95 th USAMMDA Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final percentile) were visually compared with the observed data to evaluate how well the models predicted the data.

Study 044
Based upon the modeling results in the previous publications of study 044 1 and 033 2 , a onecompartment pharmacokinetic model with first order absorption and elimination was used to describe the tafenoquine concentration versus time profiles. A schema of the onecompartment PK model is displayed in Figure 4-1.     For study 044, the population estimates of CL, V, Ka, associated inter-individual variability, residual variability and covariance between CL/F and V/F of the final model were in agreement with the published estimates.
The population predicted (PRED) plasma concentrations of tafenoquine for study 044 were obtained using model-estimated population PK parameters from the final PK model, and the published and newly modeled plots of the predicted plasma concentrations versus observed plasma concentrations of tafenoquine were presented in Figure Table 5-5 and Table 5-6 present the published and estimated Population PK parameters for study 033, respectively.   Simulations were performed using NONMEM parameter estimates from the final PK model and were carried out for the study 033 dataset. One thousand (1000) simulations for each original subject were carried out. The NONMEM control file used for the simulation of tafenoquine is presented in Attachment 3.11. Median, 5 th percentile, and 95 th percentile plot of model-predicted vs. observed concentrations is presented in Figure 5-4. Similar to the published predictive check, a majority of the tafenoquine plasma concentrations lie within the 90% CI (5 th to 95 th percentile).

USAMMDA
Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final

Discussion
Confirmatory Population PK modeling was performed on plasma tafenoquine concentrations using data published previously 1,2 . The modeling effort outlined in this report followed the modeling effort as published and presented in Table 4.1 (for study 044) and 4.2 (for study 033).
Results showed that a one-compartment PK model with first order absorption and elimination was an appropriate base PK model for describing PK of tafenoquine following oral administration.
In general, there was a good agreement between the published and modelled PK parameter and variability estimates.
Mean bootstrap estimates of all PK parameters for study 033 were compared to those from the final PK model. The bootstrap estimates agree with the final estimated PK parameter estimates as well as to the published estimates.
Simulations were performed using the NONMEM parameter estimates from the final PK model for study 033 and were carried out for actual sampling times. The results are in agreement with the published results as a majority of the tafenoquine concentrations were inside the 90% confidence interval (the 5 th and 95 th percentile lines).

Conclusions
• For both studies 044 and 033, Population PK parameters of tafenoquine and associated variability using the final model are in good agreement with the published results.

SUMMARY
Objectives: The objectives of this analysis were: • to develop population pharmacokinetic (PK) model of tafenoquine using pooled PK data from 10 clinical studies and to assess the effect of covariates such as body weight, age, sex, race, and food on the PK characteristics of tafenoquine

Methods:
This population PK analysis was performed using data obtained from 10 Phase A one-compartment PK model with first order absorption and elimination was selected to describe the PK of tafenoquine. Between-individual and residual variability terms were included in the PK model. In addition, a covariate analysis was performed to identify potential covariates affecting the PK of tafenoquine and to evaluate the extent to which the covariates accounted for the variability in the overall response. The validity of the final PK model was evaluated using bootstrapping and the posterior predictive check.
USAMMDA Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final

Results and Discussion: Pharmacokinetic
From previous modeling experience (Part 1 of this report), tafenoquine followed a monoexponential kinetics. Therefore, a one-compartment PK model with first order absorption and elimination was used as a starting point for the base PK model in the population PK analysis.
A schematic representation of the final PK model is presented in Figure 4-1. The population PK parameters of tafenoquine from the final PK model and the results from the final model evaluation by bootstrapping are presented in Table 8-1.  For oral tafenoquine, the population CL/F and V/F are determined to be 4.17 L/hr and 2470 L, respectively. The first order absorption rate constant of oral tafenoquine was 0.359 1/hr. The inter-individual variability of CL/F, V/F, and Ka was 23.6%, 24.1% and 54.1%, respectively. The final PK model revealed that apparent clearance (CL/F) of tafenoquine is a function of body weight (WT) and age. Tafenoquine CL/F was found to increase with increasing body weight and decreased with age. The final PK model revealed that apparent volume of distribution (V/F) of tafenoquine is a function of body weight (WT) and food. Tafenoquine V/F was found to increase with increasing body weight and decreased with food.
The individual predicted (IPRED) plasma concentrations of tafenoquine were obtained using the individual post hoc PK parameters from the final PK model, and the plots of the individual predicted plasma concentrations versus observed plasma concentrations of tafenoquine were presented in Conclusions: • Pharmacokinetics of oral tafenoquine follows a one-compartment model with first order absorption and elimination • Body weight and age were significant covariates with respect to CL/F of tafenoquine and CL/F was found to increase with increasing body weight and decreased with age • Body weight and food were significant covariates with respect to V/F of tafenoquine and V/F was found to increase with increasing body weight and decreased with food INTRODUCTION

OBJECTIVES
The objectives of this analysis were: • to develop population pharmacokinetic (PK) model of tafenoquine using pooled PK data from 10 clinical studies and to assess the effect of covariates such as body weight, age, sex, race, and food on the PK characteristics of tafenoquine

Primary Endpoints
• Endpoints for population PK are population estimates of PK parameters (e.g., CL/F, V/F, Ka), associated inter-subject variability and residual error.
• Endpoints for the covariate analysis are the identification of significant covariates that were retained in the PK model and quantification of their effects.

Study Design
Some key details of each of 10 studies used for this population PK modeling are given in

Description of Input Data
All data used in the population PK analysis were obtained from Army Study No's: 1, 2, 3, 4, 5, 14, 15, 33, 44, and 58. A total of 866 subjects were included in the population PK analysis.
Tafenoquine concentrations, demographic information, and clinical laboratory results from 10 studies were used to build NONMEM input data for PK analysis. Description of the PK input data is presented in End-of-Text

Handling of Input Data
The data were prepared for analysis using SAS Version 9.1.3 (SAS Institute Inc., Cary, NC). Actual dosing and actual sampling times, when available, were used for the analysis.
The following dosing data and PK sampling handling were applied to the NONMEM input data for PK analysis.
• Concentrations that were BQL or equal to LLOQ were excluded.
• Missing continuous covariates were replaced by population median value

Bioanalytical Methodology
Plasma samples were analyzed for tafenoquine concentrations using a validated liquid chromatography-tandem mass spectrometry method, with an LLOQ of either 1.00 ng/mL or 5.00 ng/mL across studies.

Data Analysis
Population PK and PKPD analyses were carried out using NONMEM Version 7. x.
Model stability based on changing of the number of significant digits specified with the estimation method and changing of the initial estimates of the parameters Potential covariates affecting the PK of tafenoquine were explored. The procedure for covariate model building is detailed in Section 11.4.3.

Description of PK Model
Based on the pharmacokinetic profile of tafenoquine from previous modeling (as described in Part 1 of this report), a one-compartment PK model with first order absorption and elimination processes was selected to describe the PK of tafenoquine. As part of the modeling process, a two-compartment PK model with first order absorption and elimination process was also tested (Attachment 7). A schema of the one-compartment PK model is displayed in Figure 4-1.
This one-compartment PK model was specified in the NONMEM control file and was parameterized in terms of CL/F, V/F, Ka using the PREDPP ADVAN2 with TRANS2 subroutine. First order conditional estimation (FOCE) with interaction between variance of inter-individual variability and the variance of residual error was used as the estimation method.

Pharmacokinetic
Between-individual variability (ETA or η), which is the difference between the individual parameter estimate and the population mean estimate of final PK parameters, was described by an exponential error model as described here:

Model for residual variability
Different structural models for residual variability were tested.

Pharmacokinetic
Residual variability (EPSILON or ε), which is the difference between the observed and the individual predicted observations of the final PK model, was described by a proportional error model as shown below:

Covariate Analysis
Covariate analysis was performed to the base PK model to identify potential covariates and to evaluate the extent to which the covariates accounted for the variability in the PK parameters.
Prior to including covariates in the population model, visual inspection of the relationship between each ETA and covariate was performed using scatter plots (End-of-Text Figure  16-21). The scatter plots were used to provide visual identification of collinearity between the covariates of interest. Covariates that were identified to demonstrate collinearity were not allowed to enter the covariate model at the same time. A decision on inclusion of covariates in the final model was also based on whether they made physiological sense. The following variables were selected for evaluation as potential covariates of CL/F, V/F, and Ka: sex, age, race, weight (WT).
For covariate selection/elimination, the following steps were followed: • Selection of the simplest structural model, to use as a valid base model, based on smallest objective function and by inspection of the pattern in the residual plots. The best estimation method, the most appropriate between-subjects variance models, and the residual error model, were identified. The resulting model was called a BASE model.
• Selection of covariates by univariate analyses, at risk 0.05 (reduction of objective function of at least 3.84). The best model for each covariate to affect each of the parameters was selected at this stage.
• Multivariate analysis: all selected covariates were added together and the model fitted to data. This reference was called FULL model.
• Backward deletion was applied until no covariate could be removed without significantly increasing the objective function, resulting in the FINAL model (likelihood ratio test at p<0.001, 1 df, objective function drop of at least 10.83).
• If a confidence interval (CI) of structural parameters included the value zero, the effect was considered not significant and the model was further simplified until all structural parameters were well estimated.
Continuous covariates in the model were centered on the population median value of the subjects/patients included in the analysis and are described in more detail in the results section.

Model Evaluation
The final PK model was evaluated using bootstrapping and a posterior predictive check.

USAMMDA
Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final

Bootstrapping
Bootstrapping is data re-sampling method for estimating sampling variances, confidence intervals, and stability of regression models. Using the bootstrap approach, the bootstrap parameter values are obtained by repeatedly fitting the final population model to a reasonable number of bootstrap samples. The mean and confidence interval (CI) values of the bootstrap parameters are then compared to the final population model parameter estimates and associated CIs from NONMEM.
Bootstrap analysis procedure consisted of the following steps: 1. The replication of the data file is obtained by random draw (with replacement) from the original data file.
2. The final model is fitted to the resulting data file, and the model parameter estimates are saved for the final analysis.

Posterior predictive check
A posterior predictive check was performed by simulating the tafenoquine concentrations from the original datasets using the parameter estimates obtained in the final PK modeling steps. One thousand (1000) predicted profiles were simulated for each original subjects/patients. Random effects were included in the simulation. The median, 5 th percentile, and 95 th percentile PK concentrations versus time profiles from the simulations were compared with observed tafenoquine concentrations.

Changes in Planned Analyses
None.

USAMMDA
Population PK Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final

Analysis Population and Data Characteristics
A total of 866 subjects were included in the population PK. A breakdown of subjects/patients by study is given in  Since oral tafenoquine displayed mono-exponential disposition in the previous population PK modeling as described in Part 1, a one-compartment PK model with first order absorption and elimination was used as a starting point for the base PK model in the population PK analysis.

Base PK Model
Based on the previous modeling experience, one-compartment PK model with first order absorption and elimination rate constants was selected as the structural model. Different error models for inter-individual and residual variability were also tested. The exponential error model was finally chosen to describe inter-individual variability of each PK parameter and the proportional error model was finally chosen to describe residual error. The base PK model was chosen based on the criteria in Section 11.3. Description of key models tested during model searches is given in Table 12-2. A two-compartment PK model was also tried but was not pursued further because unreliable estimates from bootstrap results (Attachment 7).  Once one-compartment PK model with first order absorption and elimination was selected as the structural PK model, several more models were explored for inter-individual variability and residual variability. In early attempts, when inter-individual variability terms were included for all three PK model parameters: CL, V, and Ka, p-value of mean ETABAR of Ka was statistically significant indicating that the true mean of ETA on Ka was different from 0. Before dropping the inter-individual variability term for Ka, the robustness of the model with inter-individual variability on Ka was tested using both bootstraps as well as visual predictive check plots. Since results clearly showed that inter-individual variability could be estimated with reasonable precision, a decision was made continue further modeling including interindividual variability for all 3 parameters.
As shown in Table 12-3, models with covariance between CL and V variances (Models: USARMYTAFPKB0218, USARMYTAFPKB0219, USARMYTAFPKB0220, and USARMYTAFPKB0221) resulted in the smallest OFV. Out of these 4 models, the model USARMYTAFPKB0220 had the smallest OFV. However, it was not selected as the Base PK model not only because of over-prediction of concentrations as seen in visual predictive check plots and also based on the bootstrap parameter distribution (Attachment 7). In all of the models explored, ETA shrinkage on CL and V was below 15% and was ~55% for Ka. After comparing predictive check results and bootstrap results (Attachment 5, Attachment 6, and Attachment 7), the model USARMYTAFPKB0218 was eventually selected as the Base PK model.
The population parameter estimates for the base PK model are summarized in (End-of-Text Table 16-2). NONMEM control file, output, and diagnostic plots for the base PK model are provided in Attachment 4.

Final PK Model
Covariates selection was carried out as described in Section 11.4.3. Potential covariates for the base PK model were evaluated by examining scatter plots of inter-individual variability (ETA) of PK parameter versus the covariates, the correlations among the covariates.
Description of key covariate models tested during model searches is given in Table 12-3 and  Table 12-4. Because age, body weight, RACE, SEX, and FOOD were the only covariates present for all 10 studies, these covariates were selected for covariate model exploration. Each of these covariates was included in the Base PK model (   Body weight on CL and V, age on CL, V, Ka and FOOD on V, Ka were included in the full covariate PK model (Table 12-4). Full covariate PK model was further reduced by backward elimination by dropping one covariate at a time. The results from this process of backward elimination are presented in Table 12-4. Using the criteria in Section 11.4.3, the final PK model was reached with body weight and age as covariates of CL/F and body weight and food as covariates of V/F.
The relationship of covariates with CL/F and V/F is summarized in (Table 12-5). For oral tafenoquine, the population CL/F and V/F are determined to be 4.17 L/hr and 2470 L, respectively. The first order absorption rate constant of oral tafenoquine was 0.359 1/hr. The inter-individual variability of CL/F, V/F, and Ka was 23.6%, 24.1% and 54.1%, respectively. The final PK model revealed that apparent clearance (CL/F) of tafenoquine is a function of body weight (WT) and age. These covariates accounted for 2.9% of inter-individual variability of CL/F (i.e., decrease inter-individual variability from 26.5% to 23.6%). The relationship between CL/F and both covariates was: Thus, tafenoquine CL/F was found to increase with increasing body weight and decreased with age.
The mean estimate of CL/F for an individual with a median age of 25 years in the lowest body weight (43 kg) could be as low as 0.97-fold of the average CL/F. The mean estimate of The final PK model revealed that apparent volume of distribution (V/F) of tafenoquine is a function of body weight (WT) and food. These covariates accounted for 5.5% of interindividual variability of V/F (i.e., decrease inter-individual variability from 29.6% to 24.1%).
The relationship between V/F and both covariates was: Selected diagnostic plots and end-of-text references are presented in Table 12-6. Complete results of the final PK model and diagnostic plots are in Attachment 8.

Bootstrapping
The bootstrapping technique was used to evaluate the final PK and PKPD models. One thousand (1000) bootstrap samples for PK were employed to evaluate the final model. The bootstrap estimates were derived for the PK and compared to the original parameters obtained for the final model. Results of the PK model evaluation are presented in Attachment 9. Histograms of bootstrap PK parameter estimates are displayed in Attachment 9.2.
Table 12-7 compares parameter estimates from the bootstrap to NONMEM estimates. Table  12-8 compares estimates of the variability of the random effects from the bootstrap to the corresponding NONMEM estimates. It had been proposed by Ette EI et al that if the parameter estimates from the bootstrap are within ±15% of those of final model, the parameters from the final model could be considered reliable 5,6 . In this study, the differences of mean bootstrap estimates from the NONMEM estimates of those parameters were less than 5%, demonstrating a satisfactory level of reliability of the final PK model. Overall, mean population PK parameter estimates and 95% CI obtained from the bootstrap procedure were generally comparable to the estimates and 95% CI from the final PK model.
The success rate of bootstrap runs was 100% for PK model.

Posterior Predictive Check Results
Simulations were performed using NONMEM parameter estimates from the final PK and PKPD models and were carried out for the original datasets. One thousand (1000) simulations for each original subject/patient were carried out. The NONMEM control file used for the simulation of tafenoquine is given in Attachment 10. Median, 5 th percentile, and 95 th percentile plot of model-predicted vs. observed concentrations is presented in Figure  12-1 for tafenoquine. Figure 12

SIMULATIONS
Simulation of PK data for various doses and dose regimens were carried out using the final model parameter estimates (fixed effect, random effect and residual error). In general, the simulation step included creation of data files using dummy subjects with desired sampling times and dosing regimen, running simulation with desired number of replicates using the final model output parameters (THETAs, ETAs and SIGMAs) in the control file. The output from the simulations was summarized using SAS and presented graphically in Phoenix WinNonlin. The decision on the dosing regimen to be simulated and the number of simulations to be conducted was based on the instructions provided by US Army.
Following dosing scenarios were simulated.
• A reference profile (Figure 13-1) o Three 200 mg once-daily loading doses followed by 200 mg once-weekly administration for approximately for six months • Test Profile 1 (Figure 13-2) o 200 mg once-weekly administration for approximately for six months • Test Profile 2 (Figure 13-3) o Three 100 mg once-daily loading doses followed by 100 mg once-weekly for approximately six months • Test Profile 3 (Figure 13-4) o Three 200 mg once-daily loading doses followed by 200 mg once-weekly for approximately five and half months followed by three 200 mg once-daily doses • Test Profile 4 ( Figure 13-5) o Three 200 mg once-daily loading doses followed by 200 mg once-weekly for approximately five and half months followed by two 200 mg once-daily doses • Test Profile 5 (Figure 13-6) o Three 400 mg once-daily loading doses • Test Profile 6 ( Figure 13-7) o Three 400 mg once-daily loading doses followed by 400 mg once-weekly of approximately 6 months • Test Profile 7 (Figure 13-8) o Three 400 mg once-daily loading doses followed by 400 mg once-monthly for approximately 6 months The reference profile is the dose chosen and accepted by the FDA at the end of Phase II meeting for the Phase III program. This dose level maintains trough plasma concentrations at levels > 80 ng/mL, which is believed to be the minimum concentration required to prevents breakthroughs of symptomatic malaria. Due to the potential hemolytic effect of tafenoquine in the context of G6PD screening errors, it is reasonable to contemplate eliminating the loading dose (Test Profile 1) or lowering the dose of tafenoquine (Test Profile 2) in order to determine whether alternate dosing schedule might convey the appropriate protection but improve safety. Upon return from deployment from a malaria area (in this instance at approximately six months exposure), subjects must be given tafenoquine in such a manner that plasma concentrations remain above 80 ng/mL for at least three weeks. This is to prevent against breakthroughs from Plasmodium falciparum resulting from exposure to infectious mosquitoes immediately prior to the end of deployment. The simulations compare the extension of the reference profile for three weeks post deployment versus a compressed three day dosing regimen (Test Profile 3, comparison between the two in Figure 13-9) that we refer to as reverse load. The reverse load is compared to a standard 400 mg x 3 day loading dose (Test Profile 5, comparison between the two in Figure 13-10) since it was hypothesized that the maximum plasma exposure would be similar for these regimens. A composite profile where in the reference profile is compared with Test Profile 3, Test Profile 4, Test Profile 6, and Test Profile 7 is presented in Figure 13-11. There are several studies for which safety data available for the 400 mg x 3 loading dose, which are relevant for evaluating utility of the loading dose given the similarity in plasma tafenoquine levels.
For each of the dosing scenario, 5 th (blue line), median (green line), and 95 th (red line) percentile predicted concentrations are presented with target concentration (80 ng/mL) is identified with a horizontal line on the plot.
Simulations of reference profile at various body weights are presented in Attachment 11.

Discussion
Population PK modeling was performed on pooled data from 10 Phase I/II/III studies.
A total of 866 subjects were included in this population PK modeling of tafenoquine. Consistent with earlier population PK analyses of tafenoquine, results showed that a onecompartment PK model with first order absorption and elimination was an appropriate base PK model for describing PK of tafenoquine administered orally. Inter-individual variability and residual variability were described by exponential models.
The final PK model revealed that apparent clearance (CL/F) of tafenoquine is a function of body weight (WT) and age. These covariates accounted Mean bootstrap estimates of all PK parameters were compared to those from the final PK model. The differences of mean bootstrap estimates from the NONMEM estimates of those parameters were less than 5%, demonstrating a satisfactory level of reliability of the final PK model. Posterior visual predictive checks were performed on the final PK model. These were judged to adequately reproduce drug concentrations.
Current population PK modeling of pooled data from 10 studies demonstrated consistent results with the previous independent population PK modeling for data from Study 33 and Study 44. There are some differences with the previous reports. The apparent V/F from pooled population PK modeling was 2470 L compared to 1820 L and 1110 L reported by Edstein et al. 1

Conclusions
• Pharmacokinetics of oral tafenoquine follows a one-compartment model with first order absorption and elimination • Body weight and age were significant covariates with respect to CL/F of tafenoquine and CL/F was found to increase with increasing body weight and decreased with age • Body weight and food were significant covariates with respect to V/F of tafenoquine and V/F was found to increase with increasing body weight and decreased with food USAMMDA Population PKPD Modeling Report Protocol: Tafenoquine Population PK Modeling Version: Final