Huan He (1), Kun Huang (2), Runhui Wu (2), Xiaoling Wang (1), Peng Guo (1)
(1) Clinical Research Center, Beijing Children’s Hospital, Capital Medical University, National Center for Children’s Health, Beijing, China, (2) Hematology Oncology Center, Beijing Children’s Hospital, Capital Medical University, National Center for Children’s Health, Beijing, China
Objectives: Pharmacokinetics-tailored dosing of factor VIII (FVIII) has become a popular solution for personalized prophylactic treatment in hemophilia A [1]. We have developed the population pharmacokinetic (PopPK) model of Kovaltry (BAY81-8973, a full-length recombinant factor VIII) in pediatric patients [2]. The specific PopPK models were developed based on von Willebrand factor (VWF) and blood group respectively, which both included fat-free mass (FFM). VWF was used as a covariate with the best statistical significance and substituted by blood group when VWF is not available. Previous study showed VWF could also be substituted by age [3]. However, external validation is crucial because it can determine how robust and reproducible a model is, and is considered the most stringent form of model validation [4]. Thus, the objective of our study was to compare predictive performance of all PopPK models with different substitutions by an external validation.
Methods: Pediatric patients with severe hemophilia A were enrolled. FVIII levels were measured with one-stage assay. The PopPK models were two-compartment and varied in covariate inclusion.
Model A: FFM on CL and V1, VWF on CL
Model B: FFM on CL and V1, blood group on CL
Model C: FFM on CL and V1, age on CL
Model D: FFM on CL and V1.
Predictive performance was assessed by prediction-based diagnostics firstly. Model-predicted vs observed values were graphically examined. Population predictions mean forecasting using dosing information only (scenario i). Individual predictions mean forecasting using dosing and covariate information (scenario ii). Secondly, simulation-based approach (VPC) was used. Furthermore, Bayesian forecasting was applied by using dosing, covariates and several measured factor VIII levels (scenario iii). The Bayesian predictions of FVIII at trough (time after dose≥48h) were estimated based on one, two, three or four prior concentrations respectively. Forecasting metrics included median relative predicted error (rPE) and relative root mean square error (rRMSE) describing accuracy and precision. A model was deemed acceptable when both rPE<15% and rRMSE≤30%. Each model was implemented using NONMEM v7.5.
Results: The external evaluation dataset comprised 39 patients with 228 observations. The median age, weight, and FFM of the pediatric population were 5.61years, 22.5kg, and 17.89kg respectively. The proportions of patient with blood group O and non-O were 41% and 59%. The samples were obtained at approximately 1, 3, 9, 24, 48 and 72h after dose. In the prediction-based diagnostics, the rPE of population predictions (scenario i) for model A, B, C and D were 20.84%, 7.18%, 17.50% and 19.39% (rRMSE were 68.57%, 49.89%, 63.64% and 65.25%) respectively. The inclusion of covariates (scenario ii) led to an improvement of the predictive performance in model A (rPE 11.93%, rRMSE 48.99%), C (rPE 12.78%, rRMSE 49.17%) and D (rPE 11.19%, rRMSE 45.50%) except model B (rPE 11.99%, rRMSE 51.14%). However, no model in scenario i and ii satisfied the pre-set standards. The VPC plots and model-prediction vs observation plots showed predictive performance across different substitutions were close. For trough FVIII predictions (scenario ii), the model D perform best (rPE 10.07%, rRMSE 63.25%) compared with model A (rPE 18.46%, rRMSE 76.76%), B (rPE 16.71%, rRMSE 77.26%) and C (rPE 18.91%, rRMSE 69.96%). Inclusion of a measured concentration at 24h additionally to dosing and covariate information (scenario iii vs ii) improved the predictive performance and relative predicted error was reduced from 28.39% to -6.13% on average (rRMSE 72.04% to 23.39%). For Bayesian forecasting using model A, prior information at 3h combined with 24h provided the best estimation of trough FVIII concentration (rPE -2.84%, rRMSE 24.4%) among all groups, including three or four prior observations which did not further improve the predictive ability of models. In this situation of prior information, the predictive performance of model B, C and D also displayed well with rPE -4.43%, -4.23% and 4.63% (rRMSE were 24%, 23.91% and 23.86%), respectively.
Conclusions: The PopPK models for BAY 81-8973 in pediatrics varied in their predictive performance taking different clinical scenarios (i.e. information provided) into account. The model based VWF in this external evaluation showed an acceptable predictive performance using Bayesian forecasting and VWF did not show a significant superiority.
References:
[1] Chelle P, Iorio A, Edginton AN. A personalized limited sampling approach to better estimate terminal half-life of FVIII concentrates. J Thromb Haemost. 2022; 20(9): 2012-2021.
[2] He H, Huang K, Cheng XL, et al. Development and internal validation of a clinical prediction model for individualized dosing of BAY 81-8973, a full-length recombinant factor VIII, in pediatric patients with haemophilia A, Thrombosis Research. 2023; 232: 6-14.
[3] Hajducek DM, Chelle P, Hermans C, et al. Development and evaluation of the population pharmacokinetic models for FVIII and FIX concentrates of the WAPPS-Hemo project. Haemophilia. 2020; 26(3): 384-400.
[4] Sherwin C, Kiang T, Spigarelli M, et al. Fundamentals of population pharmacokinetic modelling: validation methods. Clin Pharmacokinet. 2012; 51(9): 573-90.
Reference: PAGE 32 (2024) Abstr 10942 [www.page-meeting.org/?abstract=10942]
Poster: Drug/Disease Modelling - Paediatrics