Joga Gobburu(1) and John Lawrence(2)
(1)Office of Clinical Pharmacology and Biopharmaceutics, (2)Office of Biostatistics, Center for Drug Evaluation and Research, United States Food and Drug Administration, Rockville, MD, USA.
Introduction: One of the main objectives of the nonlinear mixed effects modeling is to provide rationale individualized dosing strategies, by explaining the inter–individual variability using intrinsic and/or extrinsic factors (covariates). Modelers rely on the asymptotic p-value generated by the software to decide whether to include or exclude a given covariate. However, no systematic evaluation of the mixed effects estimation methods and their ability to generate reliable significance levels is available in the current literature. Aim: The aim of the current study was to evaluate, using computer simulations, methods for estimating the exact significance level for including or excluding a covariate during model building.
Methods: Original data were simulated using a simple one – compartment pharmacokinetic (PK) model with (full model) or without (null model) covariates (one or two) under dense and sparse sampling schemes with either 30 or 100 subjects. The PK parameters were assumed to follow a log-normal distribution with a combined proportional and additive residual error model. The covariate values in the original data were resampled (using either permutations or parametric bootstrap methods), 1000 times, to generate data under the null hypothesis that there is no covariate effect. The original and permuted data were fitted to null and full models, using first – order (FO) and first – order condition estimation (FOCE, with or without interaction) methods, in order to compare the asymptotic and conditional p-value. Kolmogorov – Smirnov’s nonparametric test was employed to accept or reject the hypothesis that the log – likelihood ratios followed a chi-square distribution. Simulation and estimation were performed using NONMEM while SAS (ver 6.12) and S-plus 2000 were used for statistical testing and graphical display.
Results: The simulations showed that for dense data FOCE interaction yielded the best results while for the sparse data both FO and FOCE methods perform similarly. Depending on the modeling objective, the appropriate asymptotic p-value can be substituted for the conditional significance level.
Reference: PAGE 10 (2001) Abstr 173 [www.page-meeting.org/?abstract=173]
Poster: poster