Anne Kümmel (1), Jasper Dingemanse (1), Andreas Krause (1)
(1) Actelion Pharmaceuticals Ltd, Department of Clinical Pharmacology, Modeling and Simulation, Allschwil, Switzerland
Objectives: The variability of pharmacokinetic (PK) and pharmacodynamic (PD) models can drive the decision-making processes, e.g., for defining safety margins or selecting doses. Therefore, confidence of model predictions is an important aspect of their utility. Available fitting tools typically provide information about the confidence for the estimated model parameters, but information about the confidence around the primary result, e.g., the predicted concentration-time or exposure-response curve, is not available directly. The objective of this project was to develop a single-interface framework for data fitting, confidence and prediction interval derivation for any user-specified function with visualization of the results.
Methods: The methodological framework was implemented in R using available estimation methods for nonlinear closed-form models. Either nonlinear least squares or maximum-likelihood estimation is used for the estimation. Confidence intervals are calculated using either the delta method, by bootstrapping, or Monte-Carlo-simulations [Bonate 2011, Lavielle 2014]. Prediction intervals are assessed with Monte-Carlo simulations.
Results: Using two case studies, a dose-response and a PK profile, the different algorithms for confidence interval calculation are discussed with respect to implementation and results. The case studies show how the framework can be used in the study-design or analysis phase. The impact of different error models on the estimation and prediction results is demonstrated.
The utility of the framework is compared to other software typically used in PK/PD modeling that partially overlap with the presented workflow: PFIM for study design optimization, Monolix, and NONMEM/Xpose for parameter estimation and model diagnostics, and Berkeley Madonna and MLXPlore/Simulx for model simulations and visualization. These tools evaluate study designs or models based on diagnostics regarding model parameter estimates and agreement of observations and predictions or visualize simulations, but typically lack information about the confidence of the model predictions.
Conclusions: A general approach to data fitting, confidence and prediction interval calculation as well as result visualization can be implemented in an R framework, handling any user-defined closed-form function for PK/PD modeling. Different methods for estimation or confidence calculation besides different error structures are implemented, suitable for teaching and model exploration purposes.
References:
[Bonate 2011] Bonate, P.L., Pharmacokinetic-pharmacodynamic modeling and simulation. 2011, New York: Springer.
[Lavielle 2014] Lavielle, M., Mixed effects models for the population approach: Models, tasks, methods and tools. 2014: Chapman and Hall/CRC.
Reference: PAGE 24 () Abstr 3536 [www.page-meeting.org/?abstract=3536]
Poster: Methodology - Model Evaluation