Construction of model-based PK-PD software platform using R package considering compatibility of nlme structured model
Hyeseon Jeon1, Taewook Sung1, Woojin Jung1, Jung-woo Chae1,2,*, Hwi-yeol Yun1,2,*
1Department of Bio-AI convergence, Chungnam National University, DaeJeon, Republic of Korea 2College of Pharmacy, Chungnam National University, Daejeon, South Korea *Those of authors contributed equally as correspondence.
Objectives: Nowadays, with overall improvement in medical standards, demand of individualized pharmaceutical intervention is on growing. On the other hand, majority of existing applications used for clinical pharmacokinetic consultation service (CPCS) could not fully reflect the pharmacologic knowledges that modeler’s had made, and the usability of the application is going outdated. Hospitals and research institutes are also working on various models based on each clinical situation, but in order to actively utilize them in the clinical field, it is time to develop a versatile model-based platform that can integrate locally separated hospital data. In a compliance of increasing pharmacokinetic and pharmacodynamic models, this study aims to provide a mathematical model based platform that is reactive, inter-changeable, compatible to perform CPCS.
Methods: The platform is made with the programming language of R The NONMEM models presented in the published paper were collected and loaded onto the platform, and the core engine for parameter optimization, the open source libraries ‘nlmixr’ and ‘RxODE’ were used. 'RxODE' was considered as a package to solve differential equations for describing pattern models, and 'nlmixr' was considered as a package to provide algorithms for parameter estimation. When comparing aspects of nlmixr and NONMEM estimates, the golden standard of nonlinear regression analysis, similar prediction patterns were shown except for estimation time. Due to the use of open source libraries, data generation of models has the advantage of being a compatible form that can be incorporated into various analysis software in the future. As the method used for estimating the parameters of the model, first order conditional estimation with interaction(FOCEI) is designated as the default, and in some cases, the stochastic approximation expectation maximization(SAEM) and non-linear mixed effect(NLME) estimation methods can be switched by modifying the code.
Results: The built platform RCPC shows about 12% accuracy in prediction when compared to PKS, a traditional TDM software used domestically and internationally, while implementing its core functions. MAE(Mean absolute error) decreased by 0.38 (10.80%) from 3.54 to 3.16, MSE(Mean squared error) decreased by 2.87 (15.17%) from 18.94 to 16.07, RMSE(Root mean squared error) decreased by 7.27 (18.45%) from 4.35 to 4.01, and MPE(mean percentage error) decreased by 4.34 (11.03%) from 26.59 to 22.25. F20(Fraction within PE% of 20) increased by 10 (40%) from 25 to 35, and F30(Fraction within PE% of 30) increased by 15 (42%) from 35 to 50. In addition, expected time compared to PKS is short, and a model suitable for a specific patient population can be utilized.
Conclusions: The platform gave a possibility of an active model use depending on the clinical situation, and an implemented model proved itself effective in pediatric infection. In further research, the method for real-time modeling approach with more user-friendly diagnostic interface will be made onto the platform to seek for universality.
References:
[1] Pippenger, C. E., & Lesser, R. P. (1984). An overview of therapeutic drug monitoring principles. Cleveland Clinic Quarterly, 51(2), 241–254. https://doi.org/10.3949/ccjm.51.2.241
[2] Fuchs, A., Csajka, C., Thoma, Y., Buclin, T., & Widmer, N. (2013). Benchmarking therapeutic drug monitoring software: A review of available computer tools. Clinical Pharmacokinetics, 52(1), 9–22. https://doi.org/10.1007/s40262-012-0020-y
[3] Lacarelle, B., Pisano, P., Gauthier, T., Villard, P. H., Guder, F., Catalin, J., & Durand, A. (1994). Abbott PKS system: A new version for applied pharmacokinetics including Bayesian estimation. International Journal of Bio-Medical Computing, 36(1–2), 127–130. https://doi.org/10.1016/0020-7101(94)90103-1
[4] Kaschek, D., Mader, W., Kaschek, M. F., Rosenblatt, M., & Timmer, J. (2019). Dynamic modeling, parameter estimation, and uncertainty analysis in R. Journal of Statistical Software, 88(1). https://doi.org/10.18637/jss.v088.i10
[5] Bae, K. S., & Yim, D. S. (2016). R-based reproduction of the estimation process hidden behind NONMEM® part 2: First-order conditional estimation. Translational and Clinical Pharmacology, 24(4), 161–168. https://doi.org/10.12793/tcp.2016.24.4.161