A prerequisite for a well-posed inference problem is that there exists a unique solution for any given input-output relation. Such a model is said to be structurally globally identifiable. If there are several solutions or an infinite number of solutions the model is said to be structurally locally identifiable or structurally unidentifiable respectively. Structural identifiability is an important concept as parameter estimates from a structurally unidentifiable model are effectively meaningless. In practice this means that if for instance a clearance parameter CL is structurally unidentifiable, it means that the parameter CL can have any arbitrarily numerical value and the model will still be able to fit the available experimental data. For a system defined by ordinary differential equations there exist several techniques to determine whether the model is structurally identifiable or otherwise. However for mixed-effect models the identifiability concept is still very much unexplored. Because of this very little has been done on development of such techniques, despite mixed-effects modelling being a well-used modelling approach.
This work is funded through the Marie Curie FP7 People ITN European Industrial Doctorate (EID) project, IMPACT (Innovative Modelling for Pharmacological Advances through Collaborative Training).
Authors and affiliations
- David L. I. Janzén, Fraunhofer-Chalmers Centre, University of Warwick School of Engineering
- Mats Jirstrand, Fraunhofer-Chalmers Centre
- Michael J. Chappell, University of Warwick School of Engineering
- Neil D. Evans, University of Warwick School of Engineering