The nature of the full-order model
The nature of the model to be reduced is an important factor in choosing
the best method for model reduction. Linear models (e.g. most
pharmacokinetic models including physiologically based PK models) are
the easiest and work well with any method. In contrast, nonlinear
models, most pharmacodynamic QSP models, tend to be harder to simplify
and many methods need to be adapted to work with them. If the QSP model
is based on continuous functions (e.g the Bone model [52]) then most
parametric methods would be useful. However models that are a discrete
combination of large continuous functions (e.g. the coagulation model
[14]) are really only amenable to nonparametric methods.
Nonparametric methods are well suited to any type of full-order model
and appear to have no requirements about their structure.