Sylvie Retout1, France Mentré1 & René Bruno2
1) INSERM U436, CHU Pitié Salpétrière, 91 bd de lôpital, 75013 Paris, France; 2) Aventis Pharma, Département de pharmacocinétique et métabolisme, 20 av Raymond Aron, 92165 Antony cedex, France
We wished to optimise population pharmacokinetics designs under different constraints and to evaluate them by simulation for a new phase III trial of enoxaparin in patients with acute myocardial infarction. A priori values of the population parameters (three fixed effects, two random effects and one parameter for the error variance) were estimated from a first population study (448 subjects, NONMEM IV). We developed the expression of the Fisher information matrix for nonlinear mixed-effects models, including the parameter for the error variance, and used the Fedorov-Wynn algorithm to minimise its determinant. Two optimal designs and a design defined by Aventis were evaluated by simulation with NONMEM. These designs consisted of 220 subjects with 4 samples at D1 and D3 but differed by the number of elementary designs (1 to 6) and by the sampling times. The bias and the RMSE on each estimated population parameter were evaluated. The standard errors of estimation provided by NONMEM were also compared to those predicted from the inverse of the Fisher information matrix. The three designs provided precise population parameter estimates (RMSE<20%). The optimal design gave the best precision and offered a simple clinical realisation. The expected standard errors for the fixed effects were slightly different from those given by NONMEM or really observed, and were more accurate for the variance parameters.
Reference: PAGE 9 (2000) Abstr 86 [www.page-meeting.org/?abstract=86]
Poster: oral presentation