Gomeni R.(1), Mallet A.(2) and Gomeni C
1) Bestfit, Luxembourg, (2) INSERM U.436, Paris (France)
(The Non-parametric Maximum Likelihood method (NPML) has been proposed for estimating the joint distribution of the parameters of a given model using a nonparametric approach (A.Mallet, Biometrika 3:645-656, 1986). The main feature of the method is that no assumption has to be formulated about the shape of this distribution. The estimation procedure requires the specification of a measurement error model and a recent implementation allows the estimation of both fixed and random effect parameters. Biological variables which may be statistically related to the parameters (covariates) can be included in the analysis as well. No second stage models between the model parameters and these covariates need to be specified.
The Non-parametric Maximum Likelihood method is implemented in the new NPML-fit software which provides a graphical interface to organize data (input stream, data file and generated tables of results) in a hierarchical disk directories structure and to display the results of the analysis (observed vs predicted values, residuals, probability density functions, ) using 2 & 3D plots.
Specific graphical tools allow to display the joint distribution of pharmacokinetic parameters and covariates together with the probability distribution of the random effect parameters conditional on any value of the covariates and the corresponding conditional means. This allows to describe nonparametrically the relationships between pharmacokinetic parameters and covariates.
The estimated distribution of the parameters can be used as a prior distribution in a Bayesian inference to estimate individual parameters. NPML-fit includes a preprocessor allowing model specification and computational options to be defined in a user-friendly manner. The program is available in a 32-bit version (Win 95 and NT).
Reference: PAGE 6 (1997) Abstr 598 [www.page-meeting.org/?abstract=598]
Poster: oral presentation