Abstracts First North - South Conference and Workshops on Pharmacogenetics (Beating the Gene : From The Bench to the Bedside)

摘要

Population modelling seeks to describe the variability in drug behaviour both between and within subjects. On the one hand, we want to understand the behaviour of the drug, and the mechanisms that underlie it. On the other hand, we also want to use this information in a way that is maximally useful for planning initial and subsequent therapy in the next patient who comes to us who seems to be a member of that population. Parametric population modelling programmes assume that the shape of the parameter distribution in the population is either normal, lognormal, or multimodal. Most currently available software for this, such as NONMEM and the USC*PACK iterative 2 stage Bayesian programme, use either the first order (FO) or the first order, conditional expectation (FOCE) approximation to compute the likelihood or the conditional probabilities. These approximations destroy statistical consistency. Because of this, there is no guarantee that these methods will get results closer to the truth if more subjects are studied. The results may actually get worse with more subjects. Further, they are not very precise. However, there are now newer parametric population modelling programmes that do have both consistency and precision, such as the software of Lavelle in France, and the PEM programme of Leary at USC. Nonparametric (NP) population modelling programmes compute the entire most likely parameter distribution, without any constraints as to the assumption of normal, lognormal, multimodal, or any other shape. They simply obtain a discrete joint probability density that is most likely given the raw data and the error model used. Since the density is discrete, there is nothing to integrate, and there is no need for approximation. Integration is simply replaced by summation. Because of this, nonparametric modelling programmes such as NPEM and NPAG are consistent and precise. Studying more patients is guaranteed to give better results. Further, NP population models are uniquely well suited to develop maximally precise dosage regimens using the method of “multiple model” dosage design. The NP software is now capable of making any linear or nonlinear model of a drug having a single response, such as phenytoin, for example, or of the induction period of carbamazepine.