Web based platform for clinical pharmacokinetic consultation service (CPCS) with Non-linear Mixed Effect Model
Woo Jin Jung (1), Taewook Hwang (2), Sangkeun Jung (2)*, Radajka M. Savic (3)*, Jung-woo Chae (1)*, Hwi-yeol Yun (1)*
(1) College of Pharmacy, Chungnam National University, Daejeon, South Korea (2) Department of Computer Science and Engineering, Chungnam National University, Daejeon, South Korea (3) Department of Bioengineering and Theapeutic Sciences, University of California San Francisco, California, USA * Those of authors contributed equally as correspondence.
Objectives: Demand of individualized pharmaceutical intervention is on growing in certain clinical situations, yet majority of existing applications used for clinical pharmacokinetic consultation service (CPCS) could not fully reflect the pharmacologic knowledge and techniques that modelers had stacked. In a compliance of increasing pharmacokinetic and pharmacodynamic models based on non-linear mixed effect (NLME) techniques, this study aims to provide a NLME model-based platform which can deliver the model code as it is, reactive and user-friendly real-time diagnostic tool to perform CPCS.
Methods: Platform is majorly developed with the programming language of R 4.0.3 and its packages. The core libraries for parameter optimization, ‘nlmixr’ and ‘RxODE’ [1,2] was used. The model codes were written in separate R files to work as add-on documents for the platform. In consideration of compatibility with NONMEM® (ver 7.4 by ICON) which is regarded as one of the golden-standards in non-linear mixed effect analysis, platform’s estimation performance is evaluated. The real-time diagnostic methods which can help understand the status of the individual estimates were introduced into the platform. Pilot running of the platform was done on docker in the following conditions (only from 4 to 16 cores of CPU were designated in testing).
- CPU : Dual AMD Rome, 128 cores total 2.25 GHz (base), 3.4 GHz (max boost)
- GPU : NVIDIA A100 GPUs 40GB
- RAM : 1TB
Results: For pilot running of the platform, Vancomycin PK-PD model, presented by Jung et al., was implemented [3]. Being reported with the estimation performances of the package nlmixr in Schoemaker et al. [4], the nlmixr based CPCS platform is confirmed to show reasonable reproducibility in comparison with NONMEM. To apply this package in clinical situation, Bayesian forecast method is used. The individual parameters based on prior distribution was made to be re-estimated with patient's observations in the platform, then optimized PK-PD profile is generated based on newly individualized parameters. This profile is used for optimizing next dose for the patient in consult. The graphical UI structure was built with the framework of R shiny and packages of networkD3, plotly is used to provide reactivity to the visualized data. Basic VPC (visual predictive check), individual IIV (inter-individual variability) estimation status and GOF (goodness of fit) were provided for estimation diagnostics. Dose simulation and steady-state calculating function by approximation is implemented. Basic workflow of the platform is completed with addition of the package rhandsontable for model’s practical use [5].
Conclusions: The platform allows for more flexible use of drug models depending on the clinical situation like in certain subpopulations or in limited clinical data. Loading and running time for models were minimized to usable level by adjusting CPU cores dedicated. Further research for more precise, feasible, individual parameter estimation, UI modification, advanced internal diagnostic tools for clinical use, computational efficiency and linkage with model repositories like KPML (Korea Pharmacometrics Model Library) are also being considered.
Acknowledgement: This study was supported by Institute of Information and Communications Technology Planning and Evaluation grant funded by the government of the Republic of Korea (MSIT; No. 2020-0-01441, Artificial Intelligence Convergence Research Center, Chungnam National University).
References:
[1] M. Fidler et al., “Nonlinear Mixed-Effects Model Development and Simulation Using nlmixr and Related R Open-Source Packages,” CPT Pharmacometrics Syst. Pharmacol., vol. 8, no. 9, pp. 621–633, 2019, doi: 10.1002/psp4.12445.
[2] T. Odes, “R and nlmixr as a gateway between statistics and pharmacometrics,” no. November 2020, pp. 283–285, 2021, doi: 10.1002/psp4.12618.
[3] W. J. Jung et al., “Dose Optimization of Vancomycin Using a Mechanism-based Exposure–Response Model in Pediatric Infectious Disease Patients,” Clin. Ther., vol. 43, no. 1, pp. 185-194.e16, 2021, doi: 10.1016/j.clinthera.2020.10.016.
[4] R. Schoemaker et al., “Performance of the SAEM and FOCEI Algorithms in the Open-Source, Nonlinear Mixed Effect Modeling Tool nlmixr,” CPT Pharmacometrics Syst. Pharmacol., vol. 8, no. 12, pp. 923–930, 2019, doi: 10.1002/psp4.12471.
[5] http://168.188.128.119:20038/, accessed on May 12th, 2021