Comparison of nonlinear mixed effects and conventional least squares analyses of PET PK-receptor occupancy studies

Introduction: Data derived from PET studies, either at the level of individual regional time–activity curves, or at the study level, e.g. dose– occupancy curves across subjects, is conventionally estimated using non-linear least squares (NLLS). Pharmacokinetic (PK) data, on the other hand, which in contrast to PET data is sparse and from a large number of subjects, is often analysed with nonlinear mixed effect models (NLMEM). This latter approach aims to partition the total variability in the population into estimation errors and inter-subject variation. Here we compare the application of these approaches to the estimation of a simulated PET dose occupancy study, where the parameters of interest are the unoccupied specific (Vs) and non-displaceable distribution volumes (VND) of a radioligand in a given region of interest, and the occupancy by an exogenously administered drug, characterised by its plasma EC50;