Marina Cuquerella1,2,3,4, Alessandro De Carlo4, Sergio Sánchez Herrero3, Matilde Merino-Sanjuán1,2, Javier Reig-López1,2, Elena Tosca4, Victor Mangas-Sanjuán1,2, Paolo Magni4
1Department of Pharmacy and Pharmaceutical Technology and Parasitology, School of Pharmacy, University of Valencia., 2) Interuniversity Research Institute for Molecular Recognition and Technological Development, 3Simulation Department, Empresarios Agrupados Internacional S.A., 4Department of Electrical, Computer and Biomedical Engineering, University of Pavia
Introduction: Therapeutic drug monitoring represents one of the main applications of pharmacometrics in the clinical setting. Parameter estimations are routinely performed for characterizing individual pharmacokinetic (PK) parameters in order to optimize the benefit/risk balance of a patient. In spite of the limitations typically associated with the usage of Physiologically Based Pharmacokinetic (PBPK) models in this area they have demonstrated to be able of coping with limited sampling collection and reliability of predictions specially performing previously a Global Sensibility Analysis (GSA)[1]. Objectives: The main aims are: (i) implementing the Two-stage d Global Sensitivity Analysis method in PhysPK® connected with Python; (ii) assessing the impact of model parameter uncertainties on the main PK endpoints predicted with a PBPK model; (iii) evaluating iterative two-stage (ITS) method for parameter estimation with alternative sampling strategies. Methods: Regarding the semi-mechanistic PBPK model Phys-DAT [2], previously developed with PhysPK® for orally absorbed drugs (concretely for single oral 40 mg citalopram admninistration) impact of the parameter uncertainty on PK endpoints (AUC, Cmax, Tmax) was evaluated using the Two-Stage d GSA methodology [3] and considering correlations among parameters. For the most significant parameters, estimations were performed (applying ITS method [4] in Python) and different optimization algorithms (Powell, Nelder-Mead and BFGS) were compared considering six pre-established scenarios (different sampling schedules: rich or sparse and theoretical populations with individuals of p50th, p<10th, p>90th). Five or two randomly selected samples from 0 to 24 h after drug administration were considered for the rich and sparse conditions, respectively. Individuals were obtained from a virtual population of 1.000 individuals taking into account the inter-individual variability. Average Fold Error (AFE), Absolute Average Fold Error (AAFE) and Percentage Estimation Error (PEE) were considered for assessing the estimation performance. Results: GSA showed that Vd, CL and stomach emptying rate constant were the most impacting parameters in the PBPK model for predicting AUC, Cmax and Tmax PK endpoints. Estimation performance of the individual PK parameters and PK endpoints were provided. The overall results showed a better predictive performance in predicting PK endpoints rather than PK parameters. Further, it could be appreciated the impact of case-study sub-populations obtaining, in general, more extreme EE values for the worst-case scenario and the best EE values for the real case scenario. Independently on sampling richness, in real-case scenario EE values were in the range of 0.8-1.25, except for the gastric emptying rate constant and volume of distribution, which fell within the 2-fold error range. In the worst-case scenario, several AFE and AAFE values exceeded this acceptable range, particularly for the gastric emptying rate constant. The Nelder-Mead optimizer performed best in rich sampling conditions (AFE: 1.06-1.88 (Nelder-Mead); 0.57-1.89 (others); AAFE: 1.02-1.79 (Nelder-Mead);1.02-2.77 (others)). In sparse sampling conditions, Powell and Nelder-Mead produced similar results. There was a notable decline in the estimation of PK endpoints, especially for Tmax and AUC. In the best-case scenario, predictions improved, particularly AUC (AFE: 0.8-0.99; AAFE: 1.04-1.12), although the estimation of some parameters, like gastric emptying rate constant, was less accurate. The PK endpoints were generally less influenced by changes across the different scenario, with greater accuracy for both, best- and real- case situations. Tmax was the most affected, likely due to its dependence on gastric emptying rate constant variations. Conclusion: Two-Stage d GSA method offers a robust alternative for evaluating the impact of parameters on PK endpoints in presence of correlations, enhancing parameter estimation and optimizing sampling strategies. Results underscore the importance of incorporating correlations between parameters and suggest Nelder-Mead algorithm may improve estimation accuracy and precision. Influence of sampling strategies and patient characteristics is significant being crucial for enhancing precise drug dosing and maximizing pharmacokinetics reliability.
[1] Hsiesh NH, Reisfeld B, Bois FY, Chiu WA. Applying a Global Sensitivity Analysis Workflow to Improve the Computational Efficiencies in Physiologically-Based Pharmacokinetic Modeling. Front Pharmacol.2018;8;9:588 [2] Cuquerella-Gilabert M, Reig-López J, Serna J, Rueda-Ferreiro A, Merino-Sanjuan M, Mangas-Sanjuan V, Sánchez-Herrero S. Phys-DAT: A physiologically-based pharmacokinetic model for unraveling the dissolution, transit and absorption processes using PhysPK®. Comput Methods Programs Biomed. 2024 Jan;243:107929. doi: 10.1016/j.cmpb.2023.107929. Epub 2023 Nov 18. [3] De Carlo A, Tosca EM, Melillo N, Magni P. A two-stages global sensitivity analysis by using the δ sensitivity index in presence of correlated inputs: application on a tumor growth inhibition model based on the dynamic energy budget theory. J Pharmacokinet Pharmacodyn. 2023 Oct;50(5):395-409. doi: 10.1007/s10928-023-09872-w. Epub 2023 Jul 9. PMID: 37422844. [4] Magni P, Sparacino G. 5 – Parameter Estimation. In: Carson E, Cobelli C, editors. Modelling Methodology for Physiology and Medicine (Second Edition). Oxford: Elsevier; 2014.
Reference: PAGE 33 (2025) Abstr 11386 [www.page-meeting.org/?abstract=11386]
Poster: Methodology - Estimation Methods