WP7 – Simulation of clinical trials in small population groups
We will investigate the impact of model misspecifications in non-linear mixed effects models when it is used for design and analysis of trials and develop methods to mitigate it. In the design stage, clinical trial simulation can be combined with resampling procedures to arrive at more realistic trial outcomes. As a separate strategy, highly mechanistic (physiology-based pharmacokinetic and systems pharmacology) models will be combined with extended models for stochastic components (e.g. stochastic differential equations) for simulation, while more traditional pharmacometric models will be used for evaluation of expected study characteristics. For non-linear mixed effects based analyses of trials in small populations, the performance of randomisation tests and model-averaging procedures will be evaluated and methods for implementation developed.
The model uncertainty is for natural reasons largest when based on estimation in a small sample size (e.g. small population groups) and at the same time a small sample size represent an extra challenge in accurately characterising the uncertainty. Estimation of covariance-variance matrix or bootstrap resampled datasets is common methods for characterising model uncertainty with non-linear mixed effects models. Both of these methods suffer drawbacks especially in application of complex non-linear mixed effects models to studies in small population groups. We intend to demonstrate some practical limitations of these methods and develop extensions to these methodologies to improve characterisation of model uncertainty.
