Kevin Mcnally 1, Krishna Telaprolu 1, Naresh Mittapelly 1, Sebastian Polak 1, Celine Brochot 1, Masoud Jamei 1
1 Certara (Sheffield, United Kingdom)
Introduction:
Sensitivity Analysis (SA) describes a class of methods for studying the relationships between the inputs and outputs of a model [1]. It is generally expected that SA forms part of PBPK-based regulatory submission to address data gaps, uncertainty, and variability in key model parameters [2]. However quantitative global SA requires using a variance-based analysis requires many thousands of evaluations of the model under study. Even with parallelization of simulations, a conventional global sensitivity analysis (GSA) workflow is therefore impractical for computationally expensive PBPK models. A meta-modelling approach [3], whereby sensitivity analysis is conducted on a statistical approximation to the PBPK model, has the potential to reduce the computational burden of GSA by several orders of magnitude with little loss of precision, thus making variance-based GSA feasible for computationally expensive PBPK models.
Objectives:
The objective of this work was to demonstrate application of a meta-modelling-based GSA using a Gaussian Process (GP) regression model to approximate the PBPK model output. The case study was a PBPK model developed for the PERSERIS™ risperidone formulation, which is a Poly lactic-co-glycolic acid (PLGA) based in situ gel-forming implant. Due to the complexity of the underlying equations and a simulation period of several thousand hours, the model requires several minutes to complete a single evaluation.
Methods:
PBPK models of risperidone and its metabolite paliperidone were developed, characterizing the distribution and elimination of both molecules using the pharmacokinetic (PK) data of intravenous (IV) [4] and oral dosing [5, 6]. The developed models were verified using an external oral dosing dataset [7]. A model for risperidone administered as a PLGA-based in situ forming gel was subsequently developed, with polymer and implant related properties parameterised using data from a reverse-engineering study [8]. All models were implemented in the Simcyp Simulator (V25).
The influence of five uncertain parameters on the timing and magnitude of an initial peak (Tmax1, Cmax1) in plasma concentrations due to burst release, and on time and magnitude of a later peak following bulk degradation of the PLGA matrix (Tmax2, Cmax2) were studied in GSA using a meta-modelling approach. Whilst a meta-modelling approach is available in Simcyp, the custom GSA options required for this case study necessitated implementation of the GSA workflow in R. The spacefillr and sensitivity packages and some bespoke code were used for the GSA workflow, the Simcyp Simulator was used to run the PBPK models and the Simcyp-R package was used to interface between the Simcyp Simulator and R.
Results:
A total of 128 runs of the PBPK model were required for fitting and validation, with a run time of approx. one hour for simulations distributed over eight cores. GP regression models were successfully fitted to output data on Tmax1, Cmax1, Tmax2 and Cmax2. Optimization of meta models took less than one minute per summary metric. The 16,000 estimates from meta-models required for the variance-based GSA were achieved in several seconds.
Visual checks of observations against predictions supplemented by calculations of Nash–Sutcliffe efficiency suggested that an excellent fit was achieved using just 128 design points. Results from GSA identified the parameters and interactions that influenced burst release and which drove the timing and rate of bulk degradation of the PLGA matrix.
Conclusions:
The meta-modelling approach uses information in both model inputs and model outputs and consequently allows massive efficiency gains to be achieved. So long as the meta-model fit has been appropriately validated, the results using this approximation may be used as an approximation. A conventional variance-based GSA was estimated to take two weeks and was not feasible for this model due to the computational burden of simulations.
References:
[1] A. Saltelli et al. (2008). Global Sensitivity Analysis. The Primer. John Wiley & Sons (2008).
[2] EMA (2019). Guideline on the reporting of physiologically based pharmacokinetic (PBPK) modelling and simulation. EMA/CHMP/458101/2016.
[3] Oakley, J. and O’Hagan, A. (2004). Probabilistic Sensitivity Analysis of Complex Models: A Bayesian Approach. Journal of the Royal Statistical Society: Series B, 66(3).
[4] Huang, M.L., et al., Pharmacokinetics of the novel antipsychotic agent risperidone and the prolactin response in healthy subjects. Clin Pharmacol Ther, 1993. 54(3): p. 257-68.
[5] Mahdy, W.Y.B., et al., Physiologically-based pharmacokinetic model to investigate the effect of pregnancy on risperidone and paliperidone pharmacokinetics: Application to a pregnant woman and her neonate. Clin Transl Sci, 2023. 16(4): p. 618-630.
[6] Novalbos, J., et al., Effects of CYP2D6 genotype on the pharmacokinetics, pharmacodynamics, and safety of risperidone in healthy volunteers. J Clin Psychopharmacol, 2010. 30(5): p. 504-11.
[7] Borison, R.L., et al., Pharmacokinetics of risperidone in chronic schizophrenic patients. Psychopharmacol Bull, 1994. 30(2): p. 193-7.
[8] Wang, X., et al., Reverse engineering of Perseris and development of compositionally equivalent formulations. Int J Pharm, 2023. 639: p. 122948.
Reference: PAGE 34 (2026) Abstr 12283 [www.page-meeting.org/?abstract=12283]
Poster: Methodology - Model Evaluation