Y. Merlé, F. Mentré, J. Gilles and A. Mallet.
INSERM U194, GERC, Service d'Informatique médicale, CHU Pitié-Salpêtrière 75013 Paris
The Nonparametric Maximum Likelihood Method (NMPL) has been proposed for estimating the joint distribution of the parameters of a given model from retrospective data [1], The main feature of the method is that no assumption has to be formulated about the shape of this distribution. An interactive implementation of this approach has been written in Fortran and runs on Vax/VMS as well as on Unix stations.
The estimation procedure requires the specification of a structural and a measurement error model. Biological variables which may be statistically related to the parameters (the so-called 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 program provides the estimated joint distribution of the parameters and of the covariates as well as their estimated means and covariance matrix. This estimated distribution is discrete and thus in agreement with the theory of mixtures of likelihoods. From the file of results provided by NPML, graphic outputs can be obtained by using the so-called Trace-GKS program. Given the value of one or several covariates, the smoothed marginal distribution of each parameter or covariate as well as the conditional parameter distribution can be displayed. This feature is particularly useful in investigating the relationships between the parameters and the covariates.
Estimated parameter distributions of pharmacokinetic models of various drugs (gentamicin, quinidine, ciclosporine…) have already been obtained. In some cases, subpopulations of subjects have been individualized and relationships between parameters and covariates have been found [2]. Our demonstration will be based on the results obtained from one of these studies. The estimated distribution can be used in Bayesian forecasting programs in order to optimize drug dosage regimen [3].
References
1. A. Mallet. A maximum likelihood estimation method for random coefficient regression models. Biometrika 3: 645-656 (1986)
2. A. Mallet, F. Mentré, J. Gilles, A.W. Kelman, A.H. Thomson, Bryson S.M. and B. Whiting. Handling covariates in population pharmacokinetics, with an application to gentamicin. Biomed. Meas. Infor. Contr. 3: 138-146 (1988)
3. Y. Merlé, F. Mentré, A. Mallet, A. Aurengo. Computer-assisted individual estimation of radioiodine thyroid uptake in Grave’s disease. Comp. Methods Prog. Biomed. 40: 33-41 (1993).
Reference: PAGE 3 (1994) Abstr 845 [www.page-meeting.org/?abstract=845]
Poster: Software Demonstration