P. Magni(1), R. Bellazzi (1), G. De Nicolao(1), M. Simeoni(1), M. Germani(2), I. Poggesi(2), M. Rocchetti (2)
(1) Dipartimento di Informatica e Sistemistica, Pavia University, Pavia, Italy (2) Global Drug Metabolism, Pharmacia, Nerviano, Italy
Objectives: In cases in which frequent and/or fixed individual samplings cannot be applied due to ethical, technical and cost reasons, the use of the trapezoidal rule for AUC calculation is impossible or can give unreliable results. Anyway, having also in these cases a sensible estimate of the systemic exposure is of great interest. Non-linear mixed effect models can provide both population and individual estimates of AUC based on sparse sampling protocols; however, the knowledge of the structural model of pharmacokinetics is required. This could not always be available, especially in the first phases of preclinical development. In this work, we propose and evaluate a new nonparametric Bayesian scheme for AUC estimation in population studies with arbitrary sampling protocols, without the need of structural pharmacokinetic models.
Methods: The individual plasma concentration curves and the mean population curve are described by a random walk process, in which the only requirement is a certain degree of regularity of the time profiles. Population and individual AUC estimation are performed by numerically computing the posterior expectation through a Markov Chain Monte Carlo algorithm.
Results: The method was applied to simulated and real data sets. The methodology was not only able to provide good estimations of the average population AUC and its variability, but also to provide these estimation for the individual AUCs. This was possible also in cases in which samples were sparse and/or obtained at different sampling times in the different experimental units.
Conclusion: This work presents a new Bayesian approach to nonparametric AUC estimation in population studies. The method is of interest when no reliable structural model is available to describe the kinetics of the substance under analysis. The new method has wider applicability and gives better results than other nonparametric approaches (Bailer, and Yeh) and allows for arbitrary sampling and in the case of misspecified models does even better than the standard parametric approaches.
Reference: PAGE 12 (2003) Abstr 411 [www.page-meeting.org/?abstract=411]
Poster: poster