Diot Annabelle (1), Mallet Alain (1) & Bruno René (2)
(1) INSERM U436, CHU Pitié Salpetrière, 91 bd de l’Hôpital, 75013 Paris, France; (2) Genentech, Inc., South San Francisco
Several methods have been proposed in Pharmacokinetic/Pharmacodynamic population modelling for estimating random and fixed parameters in mixed models. Among non-bayesian methods, and apart from a few non-parametric methods, they all rely upon approximation of the likelihood function. The accuracy of this approximation is likely to govern the properties of the derived estimates of both the distribution of the random effects and the fixed parameters. Our hypothesis is that the subtle features of the random effects, such as multimodes, might be lost by using a rough approximation. In order to compute likelihood up to a satisfactory and adjustable degree of approximation and to be able to minimize it thereafter, we propose an approach based on the Gauss Hermite quadrature. This approach has been already proposed by others, especially in the epidemiological framework. Thanks to this quadrature, we can express the integral-based individual likelihood according to an easy to compute expression.
First, we will present this semi-parametric approach, show how it can be generalized to mixed models and how it may deal with mixtures of normal distributions for the random effects. Second, we will show its ability to estimate the population parameters of the model and to detect subpopulations with specific kinetic parameters. An application of this method is the detection of heterogeneities in groups of patients.
Simulations have been performed, based on the model and the a priori values of population parameters of a phase III clinical trial of an anticoagulant molecule. This molecule was injected by an IV bolus followed by few subcutaneous doses in 448 patients with acute myocardial infarction (Aventis Pharma, data on file). Results obtained with this method will be presented and compared with other ones.
Reference: PAGE 10 (2001) Abstr 211 [www.page-meeting.org/?abstract=211]
Poster: poster