Retout Sylvie & Mentré France
INSERM U436, CHU Pitié Salpêtrière, Paris, France
We extended the population Fisher information matrix, which was proposed to evaluate and optimise population pharmacokinetic designs. The first proposed approach used a first order Taylor expansion of the nonlinear mixed effects model around the expectation of the random effects, as in the FO method of NONMEM. We extended the development to the case of heteroscedastic variance error models and/or of additional random effects taking into account inter-occasion variability (IOV). Moreover, we added in our development the estimation of fixed parameters quantifying the relationship between individual parameters and covariates. Including those extensions, a new expression was also proposed using a linearisation around the expectation of the empirical Bayes estimates of the random effects (as in FOCE of NONMEM) using Monte Carlo integration.
We evaluated by simulation the relevance of the standard errors expected from those evaluations of the Fisher matrix. Our case study was the design of a new population pharmacokinetic study in a Phase III trial for enoxaparin, a low molecular weight heparine. Based on the results of a previous population PK study, 30 sets for 2 different population designs were simulated. The model included 5 fixed effects (2 quantifying the influence of body weight and of creatinine clearance on clearance), 3 random effects (one for IOV on clearance) and one parameter for the error variance. The standard error given by NONMEM, the empirical standard error on the 30 sets and those expected by our approach were compared.
The Fisher matrix predicted adequately the standard errors also for covariate parameters and for IOV. Furthermore, on this example, no benefit was observed between the approach using the linearisation as in FO than that in FOCE. This study confirmed that the population Fisher matrix can adequately predicts the efficiency of a design, avoiding cumbersome simulations, even for more complex population models.
Reference: PAGE 10 (2001) Abstr 195 [www.page-meeting.org/?abstract=195]
Poster: poster