Seongwon Park1, Woojin Jung2, Soyoung Lee1, Hwi-yeol Yun1,2,3, Jung-woo Chae1,2,3
1College of Pharmacy, Chungnam National University, 2Senior Health Convergence Research Center, Chungnam National University, 3Bio-AI Convergence Research Center, Chungnam National University
Objectives: Model external validation has been restricted by the limitation of limited datasets, so it has emerged as an alternative external model validation method using reported model. As meta-analysis is widely used to integrate results, this study investigates how to externally validate population pharmacokinetic (popPK) model and proposes an integrated model based on meta-analysis with the case of example of Tenofovir alafenamide (TAF) and tenofovir (TFV). Methods: Four popPK models were identified via a systematic review of PubMed [1,2,3,4] and evaluated with an in-house plasma dataset. Model evaluation included visual predictive checks (VPC, n=1000), normalized prediction distribution error (NPDE, n=1000), and prediction accuracy indices such as mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), mean percentage error (MPE), and overall prediction error distribution (PE). Based on these criteria, the best models describing TAF and TFV kinetics were selected. The $PRIOR subroutine in NONMEM was then used to constrain parameter estimation with prior information [5]. In addition, covariates from the in-house data along with those selected covariates from the model were considered, leading to the inclusion of weight and sex. The validity of prior information and appropriateness of parameters were assessed by a drop of >3.84 in the objective function value (p = 0.05) and by evaluating the relative standard error (RSE). Graphical comparisons were made by plotting the density of the assumed normal prior distributions against the posterior distributions. Model estimation and simulation were conducted using NONMEM 7.5.0 with Perl-speak-NONMEM (PsN 5.3.1), and data analysis was performed in R (v4.4.1) with RStudio (v2024.12.0+467) using ggplot2, npde, xpose4, and tidyr. Results: External validation was performed using 735 plasma TAF concentrations and 887 plasma TFV concentrations obtained after administering a single 25 mg dose of TAF to 60 healthy Asians under fasted conditions. A post-hoc analysis of the in-house data revealed that the Xingfang Ji model exhibited the lowest residuals for TAF, while the Aida N. Kawuma model showed the lowest residuals for TFV. Although NPDE analysis for all four models yielded global p-values below 0.001, the Paul Thoueille model demonstrated the best predictive performance for TAF and the Xingfang Ji model for TFV. Considering the VPC results, the integrated study selected the Xingfang Ji model (derived from healthy Chinese subjects) for TAF and the Aida N. Kawuma model (derived from HIV positive African subjects) for TFV. The Normal Inverse-Wishart Prior (NWPRI) subroutine was used to incorporate prior information into the model, constraining parameter estimation via first-order conditional estimation with interaction (FOCEI) [6]. Using the Xingfang Ji model as prior information allowed the estimation of model parameters that were well-suited to the in-house data; notably, the volume of distribution for TAF was estimated to be larger than that in the prior. Finally, posterior information was obtained by combining the in-house data with that from the Aida N. Kawuma study, and the final integrated model was evaluated using density plots and VPC. Conclusion: An integrated popPK model for TAF and TFV using meta-analysis was successfully developed and validated, proposing a model that accurately describes the pharmacokinetic profiles of healthy Asians and HIV-positive Africans. These findings suggest that the application of this methodology can enhance predictive accuracy and aid in the identification of factors influencing pharmacokinetics.
[1] Paul Thoueille et al., Population pharmacokinetic modelling to characterize the effect of chronic kidney disease on tenofovir exposure after tenofovir alafenamide administration, Journal of Antimicrobial Chemotherapy, 2023. [2] Aida N. Kawuma et al., Population Pharmacokinetics of tenofovir given as either tenofovir disoproxil fumarate or tenofovir alafenamide in an African population, CPT Pharmacometrics Syst Pharmacol, 2023. [3] Stephen A. Greene et al., Population Modeling Highlights Drug Disposition Differences Between Tenofovir Alafenamide and Tenofovir Disoproxil Fumarate in the Blood and Semen, CLINICAL PHARMACOLOGY & THERAPEUTICS, 2019. [4] Xingfang Ji et al., Population Pharmacokinetics of Tenofovir Alafenamide Fumarate and Its Metabolite Tenofovir in Healthy Chinese Volunteers, Clinical Pharmacology in Drug Development, 2024. [5] Anna H.-X. P. Chan Kwong et al., Prior information for population pharmacokinetic and pharmacokinetic/pharmacodynamic analysis: overview and guidance with a focus on the NONMEM PRIOR subroutine, Journal of Pharmacokinetics and Pharmacodynamics, 2020. [6] Per O. Gisleskog et al., Use of Prior Information to Stabilize a Population Data Analysis, Journal of Pharmacokinetics and Pharmacodynamics, 2002.
Reference: PAGE 33 (2025) Abstr 11615 [www.page-meeting.org/?abstract=11615]
Poster: Methodology - New Modelling Approaches