Van Guilder M(1), Leary R(2), Schumitzky A(1), and Jelliffe R(1)
1)Laboratory of Applied Pharmacokinetics, USC School of Medicine, Los Angeles CA,
Until now, estimates obtained in nonparametric population (NP) analyses have not included confidence intervals. To remedy this situation, we have developed a bootstrap simulation procedure to obtain them. We start with the usual NP discrete joint density, of N grid points (approximately 1 per subject studied). For each of the original subjects, we select one grid point at random. Each point is selected according to its associated probability. We then simulate the „true states” of each subject assuming parameter values given by that subject’s selected grid point. Next we perturb the „true states” for each subject with that subject’s assay and environmental error. At this point, we have created a fictive subject data set, one subject for each subject originally studied.
The NP adaptive grid (NPAG) program is then run to convergence to obtain an NP population parameter joint density. This procedure is repeated for as many population analysis simulations as desired, 1000, for example. The distributions of the various population parameter means, medians, modes, variances, correlations, and percentiles are then displayed, along with their 95 % confidence limits. This procedure now permits statements to be made about the repeatability of the results obtained with NP population PK models. The current implementation is for 3 compartment linear systems with analytic solutions. Similar software is under development for larger and nonlinear systems requiring differential equation solvers.
Supported by NIH grant LM05401.
Reference: PAGE 10 (2001) Abstr 242 [www.page-meeting.org/?abstract=242]
Poster: poster