Mats O. Karlsson, E. Niclas Jonsson, Curtis Wiltse and Janet R. Wade
Dept. of Pharmacy, Uppsala University, Uppsala, Sweden,Eli Lilly & Co, Indianapolis, Indiana, USA, The Swedish Medical Products Agency, Uppsala, Sweden
Deriving a population pharmacokinetic model from real data is always associated with numerous assumptions. Violations of these assumptions, especially if undetected, may lead to inappropriate conclusions being made from the analysis. Routinely, only a few of the assumptions are explicitly stated and justified in the reporting of a population model.
In a recent analysis of moxonidine pharmacokinetics we attempted to be exhaustive in the presentation of the assumptions made in the course of the analysis. The following list of assumptions is forming a background for a discussion regarding what (and how) may be checked during the course of an analysis.
1 Adequacy of model approximation
2 Error-free dosing history
3 Error-free sampling times
4 Error-free covariate values
5 Exclusion of some data is of no consequence
6 Invention of some data is of no consequence
7 Adequacy of the structural model
8 The same structural model applies for all subjects at all occasions
9 Adequacy of covariate identification strategy
10 Adequate shape of covariate relationships
11 Absence of covariate interactions
12 Distribution of individual parameters is appropriately accounted for
13 Heteroscedasity in variance models is appropriately accounted for
14 Adequate correlation structure of interindividual and interoccasion variability model
15 No interaction between eta’s and theta’s
16 Adequate interpretation of interoccasion parameter change as random
17 Appropriate distribution shape of residual errors
18 Independence of residual errors
19 Identical distribution of residual errors
20 The global minimum is found
21 No influencing software bugs
22 The population model can be used to generate real-world like data
Reference: PAGE 6 (1997) Abstr 605 [www.page-meeting.org/?abstract=605]
Poster: oral presentation