Lynne Evans, Leon Aarons
Department of Pharmacy, Manchester University Manchester Ml3 9PL
Sparse data arising from a population pharmacokinetic study can be analyzed by nonlinear mixed effects models. The most widely used and supported software for this is NONMEM.
A technique has been developed (Dempster et al (1), Laird and Ware(2)) based on the EM (Expectation Maximisation) algorithm for estimating variance components in linear models. This has been successfully applied to population pharmacokinetic models by Amisaki and Tatsuhara (3). In both NONMEM and the EM algorithm, population parameters are estimated by maximising the associated likelihood objective function. In an EM implementation, the fixed and random effect parameters are estimated separately, thus reducing the complexity of the maximisation. Individual parameter estimates are readily generated which can then be used to evaluate covariate relationships.
This method has been implemented and results compared with NONMEM analyses. Data used for analysis was from a drug which is currently in Phase 3/4 trials for rheumatoid arthritis. The results are very similar for ‘good’ data, however both experience problems when faced with difficult data.
[1] Dempster A.L, Laird N.M, Rubin D.B. Maximum likelihood from incomplete data via the EM algorithm. J. Roy. Stat. Soc. 39 (1977) 1-38.
[2] Laird N.M, Ware J.H. Random-effects models for longitudinal data. Biometrics 38 (1982) 963-964.
[3] Amisaki T, Tatsuhara T. An alternative two stage method via the EM algorithm for the estimation of population pharmacokinetic parameters. J. Pharmacobio-Dyn. 11 (1988) 335-348.
Reference: PAGE 3 () Abstr 863 [www.page-meeting.org/?abstract=863]
Poster: poster