Hidde van de Beek1,4, Pyry A.J. Välitalo2,3,4, J. G. Coen van Hasselt1, Laura. B Zwep1
1Division of Systems Pharmacology and Pharmacy, Leiden Academic Centre for Drug Research, Leiden University, 2School of Pharmacy, University of Eastern Finland, 3Certara Inc, 4Equal contribution
Introduction/Objectives: Population pharmacokinetic and pharmacodynamic (PK/PD) models from different studies sometimes conflict with each other [1]. When faced with conflicting models, researchers must either select which models to trust, or apply model averaging. However, model averaging can be problematic if the models are based on different study designs, or if the models do not share the same structure. To address this challenge, we have recently proposed a novel aggregate data modeling approach for pharmacometric nonlinear mixed effect models [2]. Our method allows models to be used as data, where models can be fitted using summary-level data consisting of means and covariance matrices of model parameters. Such summary level data can be derived from scientific publications, or derived from the raw data. In this presentation, we will conceptually introduce the aggregate data modelling framework for pharmacometric models. Furthermore, we will describe how an improved estimation algorithm was developed and evaluated. Finally, we will demonstrate how aggregate data modeling can be used by applying this algorithm to a meta-analysis problem would be unsolvable with other techniques. We hope with this presentation can support the use of this novel aggregate data modeling method within the broader community. Methods: Algorithm development: Since the original aggregate data MC approximation was computationally intensive, we developed an improved algorithm based on iterative reweighting MC algorithm (IR-MC), where changes in parameters result in reweighting the existing set of MC predictions. Simulation studies were performed to compare the convergence times, bias, and precision between the IR-MC algorithm and the old MC algorithm. Case study: A simulation was performed to showcase a meta-analysis problem in which data from two mono-exponential curves at different phases (distribution phase and elimination phase) are aggregated together, allowing the estimation of two-compartmental model PK parameters. A data set of 100 individuals was generated with dense sampling after 100 mg intravenous study drug dosing using a two compartmental model with inter-individual variability. Data from the PK distribution phase was used to calculate summary level data (mean vector and variance-covariance matrix), not including any elimination phase data, whereas the elimination phase data was used to fit a one-compartmental model using nlmixr2 [3]. Next, aggregate data modelling was used to fit a two-compartmental PK model to the summary data of the mono-exponential distribution phase, and to the mono-exponential elimination phase data simulated from the elimination-phase PK model. The parameter estimates were compared to a two compartmental model fit on the individual level data of both phases. Results: We demonstrate how the developed IR-MC algorithm outperforms the old MC algorithm with an 11-fold reduction in computational time, while achieving the same accuracy and precision thereby significantly improving computational efficiency. Given these improvements and to ensure broad accessibility by the community, the aggregate data modelling methods and algorithms were implemented in the “admr” R package (https://github.com/hiddevandebeek/admr). The package was designed to create a workflow for aggregate data modelling, including tasks such as calculating summary level data, deriving model-based data, and inspecting model results. The meta-analysis case study showed that the aggregate modelling fixed effects ranged from 97.5% to 110.0% of the individual data estimates. The random effects ranged from 89.3% to 123.0%. These values suggest that the aggregate data method results in highly similar fixed effects estimates, while the random effects estimates are less precisely captured. Conclusion: We have introduced and extended the aggregate data modelling framework with a more efficient algorithm, which ensures scalability of this method to more complex problems. Furthermore, the ability to combine summary-level data from different data sources and with structurally different models makes it a valuable tool for meta-analysis. These properties are unique to aggregate data modelling, distinguishing it from model-based meta-analysis and model averaging. The open-source R package “admr” ensures accessibility, allowing for future broader adoption within the pharmacometric community.
[1] Pharmacokinetic Analyses,” Clin. Pharmacokinet., vol. 51, no. 1, pp. 1–13, Jan. 2012, doi: 10.2165/11596390-000000000-00000. [2] P. A. J. Välitalo, “Pharmacometric estimation methods for aggregate data, including data simulated from other pharmacometric models,” J. Pharmacokinet. Pharmacodyn., vol. 48, no. 5, pp. 623–638, Oct. 2021, doi: 10.1007/s10928-021-09760-1. [3] M. Fidler et al., “Nonlinear Mixed-Effects Model Development and Simulation Using nlmixr and Related R Open-Source Packages,” CPT Pharmacomet. Syst. Pharmacol., vol. 8, no. 9, pp. 621–633, Sep. 2019, doi: 10.1002/psp4.12445.
Reference: PAGE 33 (2025) Abstr 11720 [www.page-meeting.org/?abstract=11720]
Poster: Methodology - New Modelling Approaches