Martin Umpierrez1, Ignacio Alvarez2, Nicolas Schmidt3, Manuel Ibarra1
1Pharmaceutical Sciences Department, Faculty of Chemistry- Universidad de la República, 2Institute of Statatistics - Universidad de la República,, 3Political Science and Quantitative Methods - Methods and Data Access Unit
Introduction: Implementation of Model-Informed Precision Dosing (MIPD) in real time at the clinical setting leverages population pharmacokinetic/pharmacodynamic (popPKPD) models to overcome limitations of traditional Therapeutic Drug Monitoring (TDM) (1). These models enable individual dose optimization first using relevant individual characteristics (covariates) for initial predictions and subsequently integrating observed concentrations and/or effects to estimate individual parameters through Bayesian updating for refined individual dosing recommendations. While popPKPD models offer significant advantages, their reliability in clinical practice requires thorough validation in the target population and specific settings where they will be applied. This validation, also called external evaluation (2–4), should be fit-for-purpose, reporting precision and accuracy of predictions with varying amount of data. Furthermore, a proper assessment for real-time MIPD should evaluate many available models identifying the most suitable one for implementation in a specific population, adding considerable complexity to the validation process. Such comprehensive comparison involves multiple sequential steps—parameter estimation, simulation, metric calculation, and summarization—for each model under consideration, creating a workflow that is often complex, time-consuming, and prone to inconsistencies when performed manually. Objetive: We aimed to develop an automated, standardized tool to streamline the external evaluation process for popPKPD models intended for real-time MIPD applications. Methods The preDose package was developed in R 4.4.2 using the usethis()(5) framework. The package supports importing popPKPD models in. mlxtran (MonolixSuite – licence required(6)) and mrgsolve (7) formats alongside external validation datasets. To implement the first, lixoftConnectors package is used (8). For the latter, mrgsolve is combined with mapbayr (9), an open-source R package. The workflow implements a sequential evaluation process: • A Priori evaluation: Assessment of model performance using only population parameters and individual covariates. •A Posteriori evaluation: Sequential Bayesian updating of individual parameters using Maximum A Posteriori (MAP) estimation as observations become available across multiple occasions. preDose offers four flexible evaluation approaches: 1.Progressive evaluation: Integrating all prior information occasion by occasion 2.Recent information approach: Using only the most recent data to predict the next occasion 3.Benchmarking against a reference occasion: Stepwise comparison against a standard reference 4.Proximity-based comparison: Using observations closest to the reference sample with sequential addition of prior information For all these, preDose automatically calculates standard predictive performance metrics (rBias, Mean Absolute Percentage Error, Root Mean Squared Error, IF20, IF30) for each sequential occasion and generates diagnostic plots using ggplot2 for model assessment and comparison Multiple models can be evaluated in parallel following the same workflow, with integrated visualization capabilities for direct performance comparison across models, facilitating objective selection of the most suitable model for implementation. The package was thoroughly tested and validated using several datasets, including real-world clinical data from different therapeutic contexts, to ensure robustness and reliability across various scenarios. Results: preDose automates the external evaluation workflow for popPKPD models, systematically quantifying how predictive performance evolves as individual data accumulates across multiple occasions. The package includes several key functionalities: a) Model and data Integration: Supports importing popPKPD models in. mlxtran and mrgsolve formats, along with datasets containing observed concentrations and covariates. b) Bayesian Update: enables the estimation of individual parameters at each occasion using the MAP approach, leveraging available observations in the external dataset. c) Flexible Evaluation: Implements the four evaluation methods described, allowing comprehensive assessment of model performance under different clinical scenarios. d) Automated Predictive Performance Assessment: Calculates standard accuracy and precision metrics for both a priori and a posteriori prediction across sequential occasions. e) Visual diagnostics: Generates comparative visualization tools that facilitate direct assessment of multiple models’ performance, supporting evidence-based model selection. By providing sequential evaluation metrics, preDose enables determination of the minimum number of observations required before model predictions become reliable enough for clinical decision-making, a critical consideration for implementing MIPD in practice. The package is available as a development version on GitHub and can be installed using: install.packages(“devtools”) devtools::install_github(“Martin-Umpierrez/R.preDose”) Conclusion: The preDose package provides a standardized, efficient, and reproducible workflow for external evaluation of popPKPD models intended for real-time MIPD applications. By automating the sequential evaluation process across multiple candidate models, preDose enables objective model selection to inform dosing decisions in the clinical setting. The ability to quantify at which point predictions become reliable enough for clinical decision-making helps bridge the gap between model development and practical implementation. This tool addresses a critical need in the field by reducing the complexity and time investment required for comprehensive model validation, ultimately facilitating the broader adoption of model-informed precision dosing in clinical practice.
1. Keizer RJ, ter Heine R, Frymoyer A, Lesko LJ, Mangat R, Goswami S. Model-Informed Precision Dosing at the Bedside: Scientific Challenges and Opportunities. Vol. 7, CPT: Pharmacometrics and Systems Pharmacology. American Society for Clinical Pharmacology and Therapeutics; 2018. p. 785–7. 2. Cheng Y, Wang C yu, Li Z ran, Pan Y, Liu M bai, Jiao Z. Can Population Pharmacokinetics of Antibiotics be Extrapolated? Implications of External Evaluations. Vol. 60, Clinical Pharmacokinetics. Adis; 2021. p. 53–68. 3. Zhao CY, Jiao Z, Mao JJ, Qiu XY. External evaluation of published population pharmacokinetic models of tacrolimus in adult renal transplant recipients. Br J Clin Pharmacol. 2016 May 1;81(5):891–907. 4. Broeker A, Nardecchia M, Klinker KP, Derendorf H, Day RO, Marriott DJ, et al. Towards precision dosing of vancomycin: a systematic evaluation of pharmacometric models for Bayesian forecasting. Clinical Microbiology and Infection. 2019 Oct 1;25(10):1286.e1-1286.e7. 5. Hadley Wickham, Jennifer Bryan, Malcolm Barrett, Andy Teucher. https://CRAN.R-project.org/package=usethis. 6. https://monolixsuite.slp-software.com/. 7. Kyle T Baron, Bill Gillespie, Charles Margossian, Devin Pastoor, Bill Denney, Dilawar Singh, et al. https://CRAN.R-project.org/package=mrgsolve. 8. https://monolix.lixoft.com/monolix-api/lixoftconnectors_installation/. 9. Felicien Le Louedec, Kyle T Baron, Laura Morvan. https://CRAN.R-project.org/package=mapbayr.
Reference: PAGE 33 (2025) Abstr 11713 [www.page-meeting.org/?abstract=11713]
Poster: Methodology - Other topics