79  When to use each method

Introduction to NONMEM 7.6.0

While there is some overlap in usage of the various EM methods, some basic guidelines may be noted. MC Importance Sampling EM (IMP) is most useful for sparse (few data points per subject, that is, fewer data points than there are etas to be estimated for a given subject) or rich data, and complex PK/PD problems with many parameters. The SAEM method is most useful for very sparse, sparse, or rich data, and for data with non-normal likelihood, such as categorical data. The iterative two stage (ITS) method is best for rich data, and rapid exploratory methods, to obtain good initial parameters for the other methods. The FOCE method is useful for rich data, and in cases where there are several or more THETAs that do not have ETA’s associated with them.